diff --git a/content/en/guide/ai-sbom_guide/1-program-foundation/1-policy/_index.md b/content/en/guide/ai-sbom_guide/1-program-foundation/1-policy/_index.md new file mode 100644 index 0000000000..0fc3e66572 --- /dev/null +++ b/content/en/guide/ai-sbom_guide/1-program-foundation/1-policy/_index.md @@ -0,0 +1,171 @@ +--- +title: "3.1 Policy" +weight: 10 +type: docs +categories: ["guide"] +tags: ["AI SBOM", "Policy"] +description: > + Explains how to establish and communicate a written policy that governs AI SBOM compliance. +--- + +{{% alert title="Implementation phase" color="info" %}} +This clause is implemented in **Phase 1: Program Foundation**. +[View the full implementation roadmap](../../#phased-implementation-roadmap) +{{% /alert %}} + +## 1. Clause Overview + +In an organization without an AI SBOM policy, developers deploy AI systems without being aware of the +license obligations of models and datasets. In AI, what must be tracked goes beyond code. Model weights, +training datasets, and the Model Tree of models derived from other models each have their own licenses, +and non-standard licenses that restrict the purpose of use, such as the Llama Community License or the +RAIL family, are common. Missing these obligations leads to copyright disputes, violations of use +restrictions, and termination of business contracts. + +To prevent this risk, 3.1 requires establishing a documented policy that governs AI SBOM compliance and +communicating it so that program participants are aware of its existence. The policy must reflect business +strategy, legal requirements in the relevant jurisdictions, and the level of risk appropriate for the use +case. All subsequent clauses (competence, license obligations, AI SBOM, governance, etc.) operate on top of +this policy. + +## 2. What to Do + +- Write and formalize a policy document that governs AI SBOM compliance. +- Define the scope of the policy (externally distributed AI systems, intake of external models and datasets, release of internal models, etc.). +- Reflect business strategy, legal requirements in the relevant jurisdictions, and the level of risk for each use case in the policy. +- Include in the policy a list of licenses allowed or prohibited for models and datasets. *([Guide recommendation] + Because compliance with non-standard licenses is difficult to track automatically after the fact, decide it by policy at the time of intake.)* +- Establish and document a procedure for communicating the policy to program participants (development, legal, security, data, etc.). +- Retain records that demonstrate the communication took place (training completion, notification history, etc.). +- Include in the policy a procedure for reviewing the policy periodically and re-communicating it when it changes. + +## 3. Requirements and Verification Materials + +| Clause | Requirement | Verification Material(s) | +|-----------|--------------|---------| +| 3.1 | A written policy shall exist that governs AI SBOM compliance, and it shall be internally communicated. The policy shall reflect business strategy, legal requirements in the relevant jurisdictions, and the level of risk appropriate for the use case. | **3.1.1** A documented policy meeting the above requirements
**3.1.2** A documented procedure that makes program participants aware of the existence of the policy (e.g., via training, internal wiki, or other practical communication method) | + +
View the original text + +> **3.1 Policy** +> A written policy shall exist that governs AI System Bill of Materials (AI SBOM) compliance. +> The policy shall be internally communicated, and informed by business strategy, legal +> requirements in the relevant jurisdictions, and the level of risk appropriate for the use case. +> +> **Verification material(s):** +> - A documented policy meeting the above requirements +> - A documented procedure that makes program participants aware of the existence of the policy +> (e.g. via training, internal wiki or other practical communication method) + +
+ +## 4. How to Comply and Samples by Verification Material + +### 3.1.1 Documented AI SBOM Policy + +**How to Comply** + +An AI SBOM policy is an official document containing the principles and procedures an organization uses to +manage the license and transparency obligations of its AI systems. The policy should include its purpose, +scope, roles and responsibilities, principles for reviewing model and dataset licenses, AI SBOM management, +response to transparency obligations, and review cycle. Since this document itself constitutes +verification material 3.1.1, manage it as an official document with its version and approval history recorded. + +An organization that already has an ISO/IEC 5230 open source policy will find it more efficient to add an +AI section to the existing policy than to create a separate new one. Add that models and datasets are +subject to license review, how non-standard licenses are handled, and in what format the AI SBOM is managed. + +A policy is not a document that is established once and left alone. Because the AI regulatory environment +changes quickly, review it at least once a year and record the change history. + +**Considerations** + +- **State the AI-specific scope**: State clearly in the policy that not only code but also model weights, + training, testing, and verification datasets, and the derivation relationships in the Model Tree are subject to license review. +- **Allowed and prohibited license lists**: Divide model and dataset licenses into an allowed list and a + prohibited list based on whether commercial use is permitted and on behavioral use restrictions, and + include them in the policy in advance. *([Guide recommendation])* +- **Approval procedure**: The legal team or the AI governance lead gives final approval, and the approval date and approver are recorded. +- **Version control**: Maintain document versions and change history so that previous versions can be compared during an audit. +- **Periodic review**: Review at least once a year, and record the review completion date and reviewer. + +**Sample** + +Below is a sample of the scope and the allowed and prohibited license lists of an AI SBOM policy. This text +becomes a core component of verification material 3.1.1. Check the actual conditions of each license against +the license text, and then classify them according to the organization's use cases. + +``` +## 1. Purpose and Scope + +This policy defines the compliance principles and procedures for the company to develop and +deploy AI systems safely and responsibly. It is designed to meet the requirements of +ISO/IEC 5230 (open source license compliance) and the OpenChain AI SBOM Compliance Guide. + +Scope: +- All AI systems, models, and services distributed externally. +- Pre-trained models and datasets brought in from outside. +- Activities that release internal models externally. + +## 2. Model and Dataset License Classification + +Licenses of models and datasets being brought in are reviewed according to the classification +below. Licenses not in the classification are reviewed by the AI governance lead before intake. + +- Allowed (commercial use permitted, no behavioral restrictions): Apache-2.0, MIT, BSD, CC-BY-4.0, etc. +- Conditionally allowed (use after review): Llama Community License, Gemma Terms of Use, + OpenRAIL family, and other licenses that restrict the purpose or scale of use +- Prohibited (non-commercial only, etc.): e.g., using a CC-BY-NC dataset in a commercial product +``` + +--- + +### 3.1.2 Policy Awareness Procedure + +**How to Comply** + +Writing a policy document is not enough. You must establish and document a communication procedure so that +program participants actually become aware of the policy's existence. This communication procedure document +itself constitutes verification material 3.1.2. Because AI systems involve not only developers but also data, +legal, and security staff, design channels so that the policy reaches all of them. + +For new hires, include an introduction to the AI SBOM policy in onboarding, and for existing employees, use +internal wiki posts and email notices. To demonstrate that the communication took place, retain evidence such +as notification history and training completion records for at least three years. + +**Considerations** + +- **Use multiple channels**: Use two or more channels, such as the internal wiki, email notices, and onboarding training. +- **Include AI-related roles**: Include dataset owners and model operators in the communication audience. +- **When the policy changes**: Have a separate procedure for notifying participants of changes immediately. +- **Retain evidence**: Retain notification history and training completion confirmations for at least three years. + +**Sample** + +Below is a sample policy communication email. Keeping a record of sending it provides evidence for verification material 3.1.2. + +``` +Subject: [AI Compliance] Notice of the AI SBOM Policy and Request to Review + +To: Employees involved in AI system development, operation, and data +From: AI Compliance Officer + +Hello, + +Our AI SBOM compliance policy has been established (or revised). +All employees involved in using, bringing in, or deploying AI models and datasets +are asked to review and familiarize themselves with the policy document below. + +- Policy document: [internal portal link] +- Key topics: model and dataset license classification, license obligation review procedure, + AI SBOM management, response to transparency obligations +- Policy version: v1.0 (effective date: YYYY-MM-DD) + +Contact: AI Compliance Officer (ai-compliance@company.com) +``` + +## 5. References + +- License obligation review procedure: [3.5 License obligations](../../2-ai-extension/1-license-obligations/) +- Basic structure of an open source policy: [ISO/IEC 5230 Conformance Guide: §3.1.1 Policy](../../../iso5230_guide/1-program-foundation/1-policy/) +- AI license management strategy: [Enterprise Open Source Management Guide: AI Compliance](../../../opensource_for_enterprise/7-ai-compliance/) diff --git a/content/en/guide/ai-sbom_guide/1-program-foundation/2-competence/_index.md b/content/en/guide/ai-sbom_guide/1-program-foundation/2-competence/_index.md new file mode 100644 index 0000000000..f4c92fce21 --- /dev/null +++ b/content/en/guide/ai-sbom_guide/1-program-foundation/2-competence/_index.md @@ -0,0 +1,136 @@ +--- +title: "3.2 Competence" +weight: 20 +type: docs +categories: ["guide"] +tags: ["AI SBOM", "Competence", "Roles"] +description: > + Explains how to define the roles and responsibilities of an AI SBOM compliance program, and how to + identify and assess the competence required for each role. +--- + +{{% alert title="Implementation phase" color="info" %}} +This clause is implemented in **Phase 1: Program Foundation**. +[View the full implementation roadmap](../../#phased-implementation-roadmap) +{{% /alert %}} + +## 1. Clause Overview + +If the policy (3.1) sets out what must be done, competence ensures there are people who can do it. AI SBOM +compliance requires broader knowledge than open source compliance, which dealt only with code licenses. +Practitioners must judge the licenses of model weights and datasets, interpret model cards, and understand +emerging AI regulations and the use restrictions of non-standard licenses. + +3.2 requires identifying the roles and responsibilities that affect the performance of the program, +determining the competence needed for each role, and ensuring that participants have that competence. +The specification states that, where relevant to the use case, participants must have competence in the +functions of governance, security, safety, privacy, development, and supplier management. + +## 2. What to Do + +- Identify and document the roles that affect the performance of the program and their responsibilities. +- Define the competence each role needs (governance, security, safety, privacy, development, supplier management). +- Add AI-specific competence (model and dataset licenses, model card interpretation, AI regulation) to the competence for each role. +- Ensure participants are competent on the basis of education, training, and experience. +- Retain evidence of competence assessments, and check periodically to keep the list up to date. + +## 3. Requirements and Verification Materials + +| Clause | Requirement | Verification Material(s) | +|-----------|--------------|---------| +| 3.2 | The organization shall identify the roles and responsibilities that affect the performance and effectiveness of the program, determine the competence each role needs, and ensure that participants have it. Where relevant to the use case, participants shall have competence in the functions of governance, security, safety, privacy, development, and supplier management. | **3.2.1** A documented list of roles with the responsibilities of each participant stated
**3.2.2** A document that identifies the competencies for each role
**3.2.3** Documented evidence of assessed competence for each participant (including periodic checks to keep the list up to date) | + +
View the original text + +> **3.2 Competence** +> The organisation shall identify the roles and the corresponding responsibilities of those roles +> that affect the performance and effectiveness of the program; determine the necessary competence +> of program participants fulfilling each role (Governance, Security, Safety, Privacy, Development, +> Supplier management if relevant to the use case); ensure that program participants are competent on +> the basis of appropriate education, training, and/or experience; and retain appropriate documented +> information as evidence of competence. +> +> **Verification material(s):** +> - A documented list of roles with corresponding responsibilities for the different participants in +> the program. +> - A document that identifies the competencies for each role. +> - Documented evidence of assessed competence for each program participant, with periodic checks to +> keep the list up-to-date. + +
+ +## 4. How to Comply and Samples by Verification Material + +### 3.2.1 List of Roles and Responsibilities + +**How to Comply** + +Document the roles involved in the program and the responsibilities of each role. In addition to general +open source roles, an AI SBOM program includes AI governance and model and dataset review roles. Stating +responsibilities specifically makes the later competence definition (3.2.2) and assignment of +responsibilities (3.8.4) clear. + +**Sample** + +``` +| Role | Responsibilities | +|------|------| +| AI governance lead | Approves the framework, judges regulatory obligations, leads periodic reviews | +| AI SBOM verification owner | Creates, reviews, and approves the AI SBOM; reflects inbound materials | +| License review owner | Judges license obligations of models, datasets, and the Model Tree | +| Data owner | Manages the provenance and licenses of training and verification datasets | +``` + +--- + +### 3.2.2 Required Competence by Role + +**How to Comply** + +Define the competence each role must have. Among the six functions listed in the specification (governance, +security, safety, privacy, development, supplier management), determine which apply to the role and the use +case, and add AI-specific competence. For example, the license review owner needs the competence to interpret +the use restrictions of non-standard licenses (Llama Community, OpenRAIL). + +**Sample** + +``` +| Role | Required competence | +|------|----------| +| AI governance lead | Governance, understanding of AI regulation (EU AI Act, Korea's AI Basic Act (AI 기본법)), risk management | +| AI SBOM verification owner | Development, SPDX and CycloneDX formats, model card interpretation, operating generation tools | +| License review owner | Supplier management, interpreting open source and non-standard licenses, Model Tree tracking | +| Data owner | Privacy, dataset licenses, provenance management | +``` + +--- + +### 3.2.3 Evidence of Competence Assessment + +**How to Comply** + +Assess whether each participant actually has the competence the role requires, and retain the evidence. +Training completion, qualifications, and practical experience serve as the basis. Because AI regulations +and licenses change quickly, check the list periodically to keep it up to date. + +**Considerations** + +- **Diversify the basis of assessment**: Use not only training completion but also work outputs (e.g., + license review records) as evidence of competence. +- **Periodic checks**: Re-examine competence requirements when new regulations take effect or new license types appear. +- **Fill gaps**: If an assessment reveals a competence gap, fill it with training or external expertise ([3.8 Resources](../../3-relevant-tasks/2-resourced/)). + +**Sample (Competence Assessment Record)** + +``` +| Participant (job) | Role | Assessment item | Basis of assessment | Result | Assessment date | +|-------------|------|----------|----------|------|--------| +| Lee OO | AI SBOM verification owner | Writing a CycloneDX ML-BOM | Internal training + work outputs | Met | 2026-03-10 | +| Park OO | License review owner | Interpreting non-standard licenses | 5 years of OSS legal experience | Met | 2026-03-10 | +``` + +## 5. References + +- Procedure for assigning resources to roles: [3.8 Effectively resourced](../../3-relevant-tasks/2-resourced/) +- Securing participant awareness: [3.3 Awareness](../3-awareness/) +- ISO/IEC 5230 competence example: [ISO/IEC 5230 Conformance Guide: §3.1.2 Competence](../../../iso5230_guide/1-program-foundation/2-competence/) diff --git a/content/en/guide/ai-sbom_guide/1-program-foundation/3-awareness/_index.md b/content/en/guide/ai-sbom_guide/1-program-foundation/3-awareness/_index.md new file mode 100644 index 0000000000..e2c6bc28e1 --- /dev/null +++ b/content/en/guide/ai-sbom_guide/1-program-foundation/3-awareness/_index.md @@ -0,0 +1,101 @@ +--- +title: "3.3 Awareness" +weight: 30 +type: docs +categories: ["guide"] +tags: ["AI SBOM", "Awareness", "Training"] +description: > + Explains how to ensure that program participants are aware of the AI SBOM policy and objectives, their + own contribution, and the implications of non-conformance. +--- + +{{% alert title="Implementation phase" color="info" %}} +This clause is implemented in **Phase 1: Program Foundation**. +[View the full implementation roadmap](../../#phased-implementation-roadmap) +{{% /alert %}} + +## 1. Clause Overview + +If competence (3.2) deals with "can they do it," awareness deals with "do they know why they must do it." +It is not enough for participants to know only that the policy exists. Compliance actually works only when +they also know how their work contributes to the program and what happens if they do not follow it. + +3.3 requires ensuring that program participants are aware of four things: the AI SBOM policy, relevant +business objectives, their own contribution to the effectiveness of the program, and the implications of +not following the program's requirements. In AI, the implications of non-compliance extend beyond copyright +disputes to regulatory violations and violations of use restrictions, so make sure participants are clearly +aware of this. + +## 2. What to Do + +- Make sure participants know about the AI SBOM policy and where to find it. +- Communicate the relevant business objectives (building trust, regulatory compliance, meeting supply chain requirements). +- Inform each participant how their work contributes to the program. +- Inform participants of the implications of non-compliance (regulatory violations, contract termination, violations of use restrictions). +- Assess participants' awareness and retain the evidence. + +## 3. Requirements and Verification Materials + +| Clause | Requirement | Verification Material(s) | +|-----------|--------------|---------| +| 3.3 | The organization shall ensure that program participants are aware of the AI SBOM policy, relevant business objectives, their own contribution to the effectiveness of the program, and the implications of not following the requirements. | **3.3.1** Documented evidence of assessed awareness for the participants, which should include the program's objectives, one's own contribution, and the implications of non-conformance | + +
View the original text + +> **3.3 Awareness** +> The organisation shall ensure that the program participants are aware of: the AI SBOM policy; +> relevant business objectives; their contribution to the effectiveness of the program; and the +> implications of not following the Program's requirements. +> +> **Verification material(s):** +> - Documented evidence of assessed awareness for the program participants, which should include: the +> program's objectives; one's contribution within the program; and the implications of program +> non-conformance. + +
+ +## 4. How to Comply and Samples by Verification Material + +### 3.3.1 Evidence of Assessed Participant Awareness + +**How to Comply** + +Assess whether participants actually understand the four elements of awareness, and keep the evidence. If +policy communication (3.1.2) proves "we informed them," the awareness assessment proves "they understood." +Assess through post-training quiz, signed acknowledgment, interviews, and similar methods. Make sure the +three elements the specification states in the verification material (the program's objectives, one's own +contribution, and the implications of non-conformance) are not left out of the assessment. + +**Considerations** + +- **Include all four elements**: In addition to policy awareness, include objectives, contribution, and the + implications of non-conformance in the assessment items. If any one is missing, it may be pointed out in a certification audit. +- **Emphasize AI-specific implications**: Include regulatory violations (EU AI Act, Korea's AI Basic Act (AI 기본법)) + and violations of non-standard license use restrictions in the implications of non-compliance. +- **Differentiate by role**: Data owners and developers contribute differently, so tailor the assessment content to each role. +- **Retain evidence**: Retain assessment results and signed acknowledgments for use as verification materials. + +**Sample (Awareness Assessment Record)** + +``` +| Participant (job) | Policy awareness | Objectives awareness | Contribution awareness | Non-conformance implications awareness | Assessment method | Assessment date | +|-------------|:--------:|:--------:|:--------:|:---------------:|----------|--------| +| Lee OO (development) | Met | Met | Met | Met | Post-training check | 2026-03-10 | +| Park OO (data) | Met | Met | Met | Met | Interview + signature | 2026-03-11 | +``` + +Example of a signed acknowledgment form: + +``` +I acknowledge that I have been informed of and understand the company's AI SBOM compliance +policy, the program's objectives, my own contribution, and the implications of non-compliance +(regulatory violations, violations of license use restrictions, contract termination). + +Name: ____ Job: ____ Signature: ____ Date: ____ +``` + +## 5. References + +- Policy and communication procedure: [3.1 Policy](../1-policy/) +- Competence by role: [3.2 Competence](../2-competence/) +- ISO/IEC 5230 awareness example: [ISO/IEC 5230 Conformance Guide: §3.1.3 Awareness](../../../iso5230_guide/1-program-foundation/3-awareness/) diff --git a/content/en/guide/ai-sbom_guide/1-program-foundation/4-scope/_index.md b/content/en/guide/ai-sbom_guide/1-program-foundation/4-scope/_index.md new file mode 100644 index 0000000000..2db18f60d1 --- /dev/null +++ b/content/en/guide/ai-sbom_guide/1-program-foundation/4-scope/_index.md @@ -0,0 +1,102 @@ +--- +title: "3.4 Program Scope" +weight: 40 +type: docs +categories: ["guide"] +tags: ["AI SBOM", "Scope"] +description: > + Explains how to clearly declare the scope and limits to which an AI SBOM compliance program applies. +--- + +{{% alert title="Implementation phase" color="info" %}} +This clause is implemented in **Phase 1: Program Foundation**. +[View the full implementation roadmap](../../#phased-implementation-roadmap) +{{% /alert %}} + +## 1. Clause Overview + +The program scope determines how far compliance applies. If the scope is vague, it becomes unclear for which +AI systems an SBOM must be created and which models' licenses must be reviewed. Only when the scope is declared +first do all subsequent clauses know what they apply to. + +3.4 requires declaring the scope that applies to each program. Scope can differ from one organization to +another. Some organizations cover a single product line, while others cover an entire department or the +entire organization. In AI, the scope decision covers not only internally developed models but also models +and datasets brought in from outside and the external release of internal models. + +## 2. What to Do + +- Determine what the program applies to (externally distributed AI systems, intake of external models and datasets, release of internal models, etc.). +- Record what is excluded and the reasons for the exclusion. +- Keep the scope statement consistent with the scope in the policy document. +- Review and update the scope periodically as the business environment changes. + +## 3. Requirements and Verification Materials + +| Clause | Requirement | Verification Material(s) | +|-----------|--------------|---------| +| 3.4 | The scope that applies shall be declared for each program. | **3.4.1** A written statement that clearly defines the scope and limits of the program | + +
View the original text + +> **3.4 Program scope** +> Different programs may be governed by different levels of scope. For example, a program could govern +> a single product line, an entire department, or an entire organisation. The scope designation needs +> to be declared for each program. +> +> **Verification material(s):** +> - A written statement that clearly defines the scope and limits of the program. + +
+ +## 4. How to Comply and Samples by Verification Material + +### 3.4.1 Program Scope Statement + +**How to Comply** + +Write a statement of the scope and limits of the program. State clearly what is covered and what is excluded, +and, for exclusions, the reasons. For an AI SBOM program, declaring what is covered by material type and by +activity makes the scope clear. The table below is an example of organizing the scope. + +**Table 1.** Example of an AI SBOM program scope declaration + +| Category | Applies | Notes | +|---|---|---| +| AI systems, models, and services distributed externally | Applies | Obligation to create an AI SBOM and review licenses | +| Pre-trained models brought in from outside | Applies | Reflected in the AI SBOM as inbound materials | +| Datasets brought in from outside | Applies | License and provenance review | +| External release of internal models | Applies | Review of the release license and transparency obligations | +| Models for internal experiments (not distributed externally) | Conditionally excluded | Applicability decided by separate review | + +**Considerations** + +- **Consistency with the policy**: The scope statement must not conflict with the scope in [3.1 Policy](../1-policy/). +- **Reasons for exclusion**: Record the reasons for excluded items. Even a model for internal experiments + comes into scope when it moves to external distribution, so have a review procedure for the point of that change. *([Guide recommendation])* +- **Periodic review**: Update the scope when a new product line or a new AI service is created. + +**Sample (Scope Statement)** + +``` +## AI SBOM Compliance Program Scope + +### In scope +This program applies to all AI systems, models, and services that the company distributes +externally, and to pre-trained models and datasets brought in from outside. It also covers +activities that release internal models externally. + +### Out of scope +Models used only for internal experiments and research and not distributed externally are +excluded. However, if such a model moves to external distribution, it goes through the intake +review procedure and is brought into scope. + +### Review cycle +The scope is reviewed and updated at least once a year as the business environment changes. +``` + +## 5. References + +- Scope of the policy: [3.1 Policy](../1-policy/) +- AI SBOM for in-scope materials: [3.9 AI SBOM](../../2-ai-extension/3-ai-sbom/) +- ISO/IEC 5230 scope example: [ISO/IEC 5230 Conformance Guide: §3.1.4 Program Scope](../../../iso5230_guide/1-program-foundation/4-scope/) diff --git a/content/en/guide/ai-sbom_guide/1-program-foundation/_index.md b/content/en/guide/ai-sbom_guide/1-program-foundation/_index.md new file mode 100644 index 0000000000..ee098be260 --- /dev/null +++ b/content/en/guide/ai-sbom_guide/1-program-foundation/_index.md @@ -0,0 +1,14 @@ +--- +title: "Program Foundation" +linkTitle: "Program Foundation" +weight: 10 +type: docs +categories: ["guide"] +tags: ["AI SBOM", "Program Foundation"] +description: > + The phase that lays the foundation of the AI SBOM compliance program. It establishes policy, competence, awareness, and scope. +--- + +This is Phase 1 of the implementation roadmap. It defines the scope of the program (3.4), establishes the +policy (3.1), and secures the competence and awareness of participants (3.2, 3.3). All subsequent clauses +operate on this foundation. diff --git a/content/en/guide/ai-sbom_guide/2-ai-extension/1-license-obligations/_index.md b/content/en/guide/ai-sbom_guide/2-ai-extension/1-license-obligations/_index.md new file mode 100644 index 0000000000..6c9e35877a --- /dev/null +++ b/content/en/guide/ai-sbom_guide/2-ai-extension/1-license-obligations/_index.md @@ -0,0 +1,153 @@ +--- +title: "3.5 License Obligations" +weight: 10 +type: docs +categories: ["guide"] +tags: ["AI SBOM", "License", "Model Tree"] +description: > + Explains the procedure for reviewing the licenses of an AI system's code, weights, datasets, and Model + Tree to determine obligations, restrictions, and rights. +--- + +{{% alert title="Implementation phase" color="info" %}} +This clause is implemented in **Phase 2: AI Extension Processes**. +[View the full implementation roadmap](../../#phased-implementation-roadmap) +{{% /alert %}} + +## 1. Clause Overview + +License obligations are where the AI SBOM guide extends ISO/IEC 5230 the most. Whereas traditional open +source compliance reviewed the licenses of code, in AI the review targets grow to four: the AI system's +code, model weights, datasets (including training, testing, and verification datasets), and the license of +the AI system itself. Because it is common for one model to be derived from several other models, each +upstream model in the Model Tree has its own license. + +3.5 requires a procedure for reviewing these licenses and determining the obligations, restrictions, and +rights that each license grants, in light of the intended use of the AI system. The review covers both the +obligations inherited from upstream and the obligations passed on downstream. + +## 2. What to Do + +- Establish a procedure for identifying the licenses of code, weights, datasets, and the AI system itself. +- Track the license of each upstream model in the Model Tree, and document the obligations, restrictions, and rights of each license. +- Use automated scanning tools for the first identification of source code and dependencies. *([Guide recommendation])* +- Judge model weights, datasets, and non-standard licenses through legal or governance review. *([Guide recommendation])* +- Have an intake procedure that requires license metadata to be received along with any model or dataset brought in from outside. + *([Guide recommendation])* +- Record the review results (obligations, restrictions, rights) in the AI SBOM and track them. + +## 3. Requirements and Verification Materials + +| Clause | Requirement | Verification Material(s) | +|-----------|--------------|---------| +| 3.5 | A procedure shall exist for reviewing the licenses of an AI system's code, weights, and datasets, as well as the license of the AI system itself, to determine the obligations, restrictions, and rights granted by each license, taking into account the intended use. Note that the upstream models in the Model Tree may each have their own license. | **3.5.1** A documented procedure to review and document, as appropriate, the upstream and downstream obligations, restrictions, and rights granted by each identified license | + +
View the original text + +> **3.5 License obligations** +> A process shall exist for reviewing the relevant identified licenses for an AI system's code, +> weights, and datasets (including but not limited to training, testing, and verification datasets) +> as well as the license for the AI system itself to determine the obligations, restrictions, and +> rights granted by each license, taking into account the intended use of the AI system. Note that +> it's often the case that an AI system is trained on multiple other AI systems that may be +> identified in the AI system Model Tree for example; each of these may have their own licenses. +> +> **Verification material(s):** +> - A documented procedure to review and document upstream and downstream obligations, +> restrictions, and rights granted by each identified license, as appropriate. + +
+ +## 4. How to Comply and Samples by Verification Material + +### 3.5.1 Procedure for Reviewing and Documenting License Obligations + +**How to Comply** + +Design the review procedure on the premise that the level of automation differs by material type. Source +code and dependency licenses can be identified to a large extent with automated scanning tools such as +FOSSology, ScanCode, and OSS Review Toolkit. However, the licenses of model weights and datasets, and the +derivation relationships in the Model Tree, are outside the coverage of these tools or are identified with +low accuracy. Restrictions on the purpose of use in non-standard licenses require human interpretation. +A practical division of work is therefore to do the first identification with automated scanning and to +pass models, datasets, and non-standard licenses to legal or governance review. + +The figure below shows the decision flow for determining license obligations when materials come in. + +```mermaid +flowchart TD + A([AI system material input]) --> B{Material type} + B -->|Code and dependencies| C[Automated SCA scan\nScanCode, ORT] + B -->|Models and datasets| D[Check license metadata\nmodel card, datasheet] + C --> E{License identified?} + D --> E + E -->|Standard allowed license| F[Classify obligations,\nrestrictions, rights] + E -->|Non-standard or unclear| G[Legal and governance review\ninterpret use restrictions] + G --> F + F --> H{Conflicts with\nintended use?} + H -->|Yes| I[Mitigate risk or\nreplace material] + H -->|No| J[Record in AI SBOM\nupstream and downstream obligations] + I --> J + J --> K([Periodic re-review]) + + style A fill:#2d3748,color:#fff + style G fill:#744210,color:#fff + style I fill:#c53030,color:#fff + style J fill:#2b6cb0,color:#fff +``` + +**Figure 1.** Decision flow for reviewing license obligations + +**Considerations** + +- **Enforce metadata at the intake gate**: The biggest cause of missed license obligations is license drift, + in which provenance and license information is lost as a model is passed downstream. One study reports + that a large share of restrictive clauses disappear in the transition from model to application + ([arXiv:2509.09873](https://arxiv.org/abs/2509.09873)). Rather than trying to restore the information + downstream, it is more effective to block intake of materials without license metadata at the internal + model and dataset registry. *([Guide recommendation])* +- **Decide on non-standard licenses in advance through policy**: Compliance with the behavioral use + restrictions of the Llama Community License or the OpenRAIL family is difficult to track automatically + after the fact. Decide at the time of intake using the allowed and prohibited lists in + [3.1 Policy](../../1-program-foundation/1-policy/). *([Guide recommendation])* +- **Track the Model Tree**: Check through the model card which upstream models the incoming model was + derived from, and review whether the license obligations of upstream models are passed downstream. +- **Check dataset use restrictions**: Check whether datasets under non-commercial licenses such as + CC-BY-NC were used to train commercial products. Missing and incorrect dataset licenses are common, so + compare against the license text. +- **Recognize the limits of automation**: Do not treat the output of automated scanning tools as the final + judgment. Tools help with identification, and people handle interpreting obligations and judging conflicts. + +**Sample** + +Below is a sample of the core part of a license obligation review procedure. This procedure document becomes verification material 3.5.1. + +``` +## AI License Obligation Review Procedure + +### 1. Review targets +- AI system code and dependencies +- Model weights (models brought in, fine-tuned models) +- Datasets (training, testing, verification) +- Licenses of upstream models in the Model Tree + +### 2. Review steps +1) Automated identification: Scan the licenses of code and dependencies with an SCA tool. +2) Metadata collection: Collect the licenses of models and datasets from model cards and datasheets. + If metadata is missing, put the intake on hold. +3) Classification: Compare the identified licenses against the policy's allowed, conditional, and prohibited lists. +4) Legal review: For non-standard or unclear licenses, interpret the obligations through legal and governance review. +5) Recording: Record upstream and downstream obligations, restrictions, and rights in the AI SBOM. + +### 3. Responsibilities and frequency +- First identification: development owner +- Interpretation of obligations: legal and AI governance lead +- Re-review: when a model or dataset is replaced, and at least once a quarter +``` + +## 5. References + +- Allowed and prohibited license list policy: [3.1 Policy](../../1-program-foundation/1-policy/) +- AI SBOM in which to record review results: [3.9 AI SBOM](../3-ai-sbom/) +- Automated scanning tools: [Tools: FOSSology](../../../tools/1-fossology/), [SCANOSS](../../../tools/9-scanoss/) +- AI model and dataset license strategy: [Enterprise Open Source Management Guide: AI Compliance](../../../opensource_for_enterprise/7-ai-compliance/) diff --git a/content/en/guide/ai-sbom_guide/2-ai-extension/2-transparency-obligations/_index.md b/content/en/guide/ai-sbom_guide/2-ai-extension/2-transparency-obligations/_index.md new file mode 100644 index 0000000000..1ef57a4b0c --- /dev/null +++ b/content/en/guide/ai-sbom_guide/2-ai-extension/2-transparency-obligations/_index.md @@ -0,0 +1,150 @@ +--- +title: "3.6 Transparency Obligations" +weight: 20 +type: docs +categories: ["guide"] +tags: ["AI SBOM", "Transparency", "AI Regulation"] +description: > + Explains the procedure for reviewing transparency obligations imposed by regulations and taking risk + mitigation measures for issues such as disclosure of training data. +--- + +{{% alert title="Implementation phase" color="info" %}} +This clause is implemented in **Phase 2: AI Extension Processes**. +[View the full implementation roadmap](../../#phased-implementation-roadmap) +{{% /alert %}} + +## 1. Clause Overview + +If license obligations (3.5) ask "do we have the right to use this material," transparency obligations ask +"what must we disclose about this material." The two obligations come from different sources. License +obligations are imposed by copyright holders through contract, and transparency obligations are imposed by +regulations through law. + +3.6 requires a procedure for reviewing whether there are transparency obligations imposed by regulations. +The review covers training, testing, and verification datasets and takes into account the intended use of +the model. If the use case for the training data creates a transparency issue (e.g., disclosure obligations +to downstream recipients), appropriate risk mitigation measures must be taken. As the EU AI Act begins to +fully apply its transparency obligations from August 2026, this clause is becoming more important in practice. + +## 2. What to Do + +- Have a procedure for identifying the transparency regulations that apply to the AI systems you bring in or develop. +- Review, based on the intended use, whether disclosure obligations apply to training, testing, and verification datasets. +- If there are disclosure obligations to downstream recipients, decide on risk mitigation measures. +- Document the transparency measures taken. +- Periodically update and reflect the latest transparency obligations set by regulators. *([Guide recommendation])* + +## 3. Requirements and Verification Materials + +| Clause | Requirement | Verification Material(s) | +|-----------|--------------|---------| +| 3.6 | A procedure shall exist for reviewing whether there are transparency obligations imposed by regulations, including for training, testing, and verification datasets, taking into account the intended use of the model. If there is a transparency issue (e.g., downstream disclosure obligations), risk mitigation measures should be taken. | **3.6.1** A documented procedure to review and document the transparency measures undertaken | + +
View the original text + +> **3.6 Transparency obligations** +> A process shall exist for reviewing if there are any transparency obligations from regulations +> including but not limited to training, testing, and verification datasets, taking into account the +> intended use of the model. If the use case for the training data creates a relevant issue (e.g., +> disclosure obligations to downstream recipients) in the context of transparency, then appropriate +> risk mitigation measures should be undertaken. +> +> **Verification material(s):** +> - A documented procedure to review and document the transparency measures undertaken. + +
+ +## 4. How to Comply and Samples by Verification Material + +### 3.6.1 Procedure for Reviewing and Documenting Transparency Obligations + +**How to Comply** + +Because transparency obligations differ by regulation, start by identifying which regulations apply. Once the +applicable regulations are determined, derive the disclosure items each regulation requires, and reflect those +items in the AI SBOM or model card. Unlike license obligations, transparency obligations center on +"disclosure," so the deliverables must be organized in a form that can be delivered externally. + +The table below lists the main transparency obligations that intersect with the AI SBOM. The regulatory +timeline and overall context are managed together in the regulatory matrix in +[3.10 Governance](../../4-governance/1-governance/). + +**Table 1.** Transparency obligations that intersect with the AI SBOM (as of 2026-06) + +| Source | Transparency obligation | Reflected in the AI SBOM and model card | +|---|---|---| +| EU AI Act Article 53 (GPAI) | Publishing a summary of training data, respecting copyright opt-outs | Dataset provenance and license, records of opt-out handling | +| EU AI Act Article 50 | Labeling AI-generated content, notice of interaction with AI | Output labeling policy | +| Korea's AI Basic Act (AI 기본법) | Labeling obligations for high-impact and generative AI, disclosure of training data provenance | Labeling and provenance items in the model card | +| Notices required by licenses | "Built with Llama" notice, stating the derived model name, etc. | Tracked together with license obligations (3.5) | + +The following figure shows the review flow for deriving transparency obligations from the intended use of the material. + +```mermaid +flowchart TD + A([Models, datasets, and intended use]) --> B[Identify applicable regulations\nEU AI Act, Korea's AI Basic Act, etc.] + B --> C{Transparency\nobligation exists?} + C -->|No| D[Record review result\nreason for no obligation] + C -->|Yes| E[Derive disclosure items\ndata summary, labeling, provenance] + E --> F{Downstream\ndisclosure issue?} + F -->|Yes| G[Risk mitigation measures\nsummary disclosure, contractual notice] + F -->|No| H[Reflect in AI SBOM and model card] + G --> H + H --> I([Document measures\nperiodic update]) + + style A fill:#2d3748,color:#fff + style E fill:#2b6cb0,color:#fff + style G fill:#744210,color:#fff + style I fill:#22543d,color:#fff +``` + +**Figure 1.** Transparency obligation review flow + +**Considerations** + +- **Dataset provenance is central**: Many transparency obligations apply to training data. Disclosure + obligations can be fulfilled only if the provenance and license of datasets are recorded in the AI SBOM. + This connects directly to the AI SBOM (3.9). +- **Intended use is the criterion**: Even for the same model, transparency obligations differ by use case. + High-risk uses and services for the general public carry heavier obligations. +- **Downstream disclosure obligations**: When supplying a model or system externally, review whether there is + information that must be given to the recipient. Risk mitigation can be carried out by publishing a summary + of training data or through contractual notice. +- **Reflect regulatory changes**: The EU AI Act applies transparency obligations from August 2026, so update + the procedure in line with that timing. Responsibility for updates is managed by governance (3.10). + +**Sample (Transparency Obligation Review Procedure)** + +Below is a sample of the core part of a transparency obligation review procedure. This procedure document becomes verification material 3.6.1. + +``` +## Transparency Obligation Review Procedure + +### 1. Identify applicable regulations +Identify the applicable regulations based on the intended use and deployment region of the AI system. +(e.g., deployment in the EU market: EU AI Act; high-impact AI in Korea: Korea's AI Basic Act) + +### 2. Derive disclosure items +Organize the transparency obligations of each regulation into disclosure items. +- Summary of training data (EU AI Act Article 53) +- Labeling of AI generation and interaction (EU AI Act Article 50, Korea's AI Basic Act) +- Disclosure of data provenance (Korea's AI Basic Act) + +### 3. Downstream review +When supplying externally, review the information to be delivered to recipients, and decide on the necessary risk mitigation measures. + +### 4. Reflection and documentation +Reflect the derived disclosure items in the AI SBOM and model card, and record the measures taken. + +### 5. Responsibilities and frequency +- Review: legal and AI governance lead +- Update: when the regulatory enforcement timeline changes, and at least once every six months +``` + +## 5. References + +- Distinction from license obligations: [3.5 License obligations](../1-license-obligations/) +- AI SBOM that holds the disclosure items: [3.9 AI SBOM](../3-ai-sbom/) +- Regulatory timeline and governance: [3.10 Governance](../../4-governance/1-governance/) +- AI model licenses and labeling obligations: [Enterprise Open Source Management Guide: AI Compliance](../../../opensource_for_enterprise/7-ai-compliance/) diff --git a/content/en/guide/ai-sbom_guide/2-ai-extension/3-ai-sbom/_index.md b/content/en/guide/ai-sbom_guide/2-ai-extension/3-ai-sbom/_index.md new file mode 100644 index 0000000000..1d35816b60 --- /dev/null +++ b/content/en/guide/ai-sbom_guide/2-ai-extension/3-ai-sbom/_index.md @@ -0,0 +1,249 @@ +--- +title: "3.9 AI SBOM" +weight: 30 +type: docs +categories: ["guide"] +tags: ["AI SBOM", "SPDX", "CycloneDX", "ML-BOM"] +description: > + Explains the procedure and formats for creating and managing an AI SBOM, automation tools, and the + verification areas that tools cannot easily cover. +--- + +{{% alert title="Implementation phase" color="info" %}} +This clause is implemented in **Phase 2: AI Extension Processes**. +[View the full implementation roadmap](../../#phased-implementation-roadmap) +{{% /alert %}} + +## 1. Clause Overview + +An AI SBOM (AI System Bill of Materials) is a list of the components that make up an AI system and +information about them. Whereas a traditional SBOM records software components, an AI SBOM adds models, +weights, datasets, and hyperparameters. 3.9 requires a procedure for creating and managing an AI SBOM. + +The format is not fixed. The specification states that any format, such as SPDX, CycloneDX, or another +format, is acceptable. There is, however, one obligation. The AI SBOM shall account for inbound materials +from third parties. If pre-trained models and datasets brought in from outside are missing, there is no +basis for tracking license obligations ([3.5](../1-license-obligations/)) and transparency obligations. + +The AI SBOM area is where the boundary "generation is automated by tools, but accuracy and compliance +judgments are filled in by people" is most visible. This page explains the procedure along that boundary. + +## 2. What to Do + +- Establish a procedure for identifying, tracking, reviewing, approving, and archiving the components of an AI system (models, datasets, etc.). +- Decide on an AI SBOM format (SPDX 3.0 AI Profile or CycloneDX ML-BOM recommended). *([Guide recommendation])* +- Always include models and datasets coming in from third parties in the AI SBOM. +- Integrate generation tools with CI/CD to generate the AI SBOM repeatedly. *([Guide recommendation])* +- Have people review whether the license and provenance fields of the generated AI SBOM are accurate. *([Guide recommendation])* +- Retain records that demonstrate the procedure was followed (generation history, approval history). + +## 3. Requirements and Verification Materials + +| Clause | Requirement | Verification Material(s) | +|-----------|--------------|---------| +| 3.9 | A procedure shall exist for creating and managing an AI SBOM. Any format, such as SPDX or CycloneDX, is acceptable. The AI SBOM shall account for inbound materials from third parties. | **3.9.1** A documented procedure for identifying, tracking, reviewing, approving, and archiving information related to the components of an AI system (models, datasets, etc.)
**3.9.2** Records for the supplied system that demonstrate the procedure was properly followed | + +
View the original text + +> **3.9 AI System Bill of Materials** +> A process shall exist for creating and managing an AI SBOM, this can be in any format e.g. SPDX, +> CycloneDX, or another format. The AI SBOM shall account for inbound materials from third-parties. +> +> **Verification material(s):** +> - A documented procedure for identifying, tracking, reviewing, approving, and archiving +> information related to the components of an AI system (e.g., model, datasets, etc). +> - Records for the supplied system that demonstrates the documented procedure was properly followed. + +
+ +## 4. How to Comply and Samples by Verification Material + +### 3.9.1 AI SBOM Management Procedure (Identifying, Tracking, Reviewing, Approving, Archiving) + +**How to Comply** + +Design the AI SBOM procedure in four stages: generation, review, approval, and archiving. Automate the +generation stage with tools, and have people handle the review and approval stages. Even if a tool copies +the license field of a model card as is, the tool cannot judge whether that license fits the actual use +case or whether anything is missing or incorrectly stated. + +The figure below shows the flow from AI SBOM generation to archiving. + +```mermaid +flowchart TD + A([AI system materials]) --> B[Generate: run SBOM tool\ncdxgen, OWASP AIBOM Generator] + B --> C[Output format\nSPDX 3.0 AI Profile\nor CycloneDX ML-BOM] + C --> D{Inbound materials\nincluded?} + D -->|Missing| E[Collect additional third-party\nmodels and datasets] + E --> C + D -->|Included| F[Review: human check of\nlicense and provenance fields] + F --> G{Accurate and complete?} + G -->|No| H[Supplement or correct] + H --> F + G -->|Yes| I[Approve: governance lead] + I --> J([Archive: version and history management\nvulnerability monitoring]) + + style A fill:#2d3748,color:#fff + style B fill:#2b6cb0,color:#fff + style F fill:#744210,color:#fff + style J fill:#22543d,color:#fff +``` + +**Figure 1.** Procedure from AI SBOM generation to archiving + +**Tool Mapping** + +These are open source tools that can be used at each stage. "Automation level" indicates how much of the +task a tool can take over. + +| Stage | Task | Automation Level | Representative Tools | +|------|------|------------|-----------| +| Generation | Code and dependency BOM | Mature | cdxgen, Syft | +| Generation | Model and metadata AIBOM | Tools emerging | OWASP AIBOM Generator, cdxgen `aibom` | +| Analysis | Static inspection of model binaries | Tools emerging | Lab700x AI SBOM Scanner | +| Management | SBOM storage and vulnerability monitoring | Mature | Dependency-Track, SW360 | +| Review | Judging license and provenance accuracy | People and policy | Tools at a supplementary stage | + +Installation, usage, and screenshots of each tool are covered in detail in the [Tools](../../5-tools/) section ([OWASP AIBOM +Generator](../../5-tools/1-aibom-generator/), [cdxgen](../../5-tools/2-cdxgen/), +[Model and container scanners](../../5-tools/3-scanners/)). + +The commands to generate an AI BOM with cdxgen are as follows. Hugging Face model URLs and purls, Modelfiles, +and GGUF artifacts can be given directly as input ([cdxgen AI-BOM documentation](https://github.com/cdxgen/cdxgen/blob/master/docs/AI_BOM.md)). + +```bash +# Generate an AI BOM in the AI project directory +cdxgen -t ai -o aibom.json . + +# Generate including AI/ML metadata (formulation) +cdxgen -t ai --include-formulation -o aibom.json . +``` + +OWASP AIBOM Generator takes a Hugging Face model as input, creates an AIBOM in CycloneDX format, and gives it a +completeness score. It is maintained by the OWASP Gen AI Security Project and is also available as a Hugging Face Space +([OWASP AIBOM Generator](https://genai.owasp.org/resource/owasp-aibom-generator/)). + +**Hands-on: Generating with cdxgen** + +This is the result of actually running cdxgen on a summarization app (depending on `transformers` and `torch`) +that loads a pre-trained model (`facebook/bart-large-cnn`). The tool automatically identifies 5 dependencies +and creates a BOM in CycloneDX 1.7 format. + +```text +$ cdxgen -t python --include-formulation -o aibom.json . +CycloneDX Generator 12.5.1 (Node.js) + +Generated components: 5 (CycloneDX 1.7): + transformers 4.44.2 pkg:pypi/transformers@4.44.2 license: empty + torch 2.4.0 pkg:pypi/torch@2.4.0 license: empty + numpy 1.26.4 pkg:pypi/numpy@1.26.4 license: empty + tokenizers 0.19.1 pkg:pypi/tokenizers@0.19.1 license: empty + huggingface-hub 0.24.6 pkg:pypi/huggingface-hub@0.24.6 license: empty +``` + +One component of the generated BOM is shown below. The identification basis (evidence) is filled in, but the +`licenses` field is empty. + +```json +{ + "name": "transformers", + "version": "4.44.2", + "purl": "pkg:pypi/transformers@4.44.2", + "type": "library", + "evidence": { + "identity": [ + { "field": "purl", "confidence": 0.5, + "methods": [{ "technique": "manifest-analysis", "value": "requirements.txt" }] } + ] + } +} +``` + +**Figure 2.** cdxgen 12.5.1 output *(run on 2026-06-13, `-t python --include-formulation`)* + +{{% alert title="What the output shows: generation by tools, verification by people" color="warning" %}} +- The tool automatically identified 5 dependencies from `requirements.txt` and created a BOM. Generation is automated. +- However, the `licenses` field of each component is empty. License accuracy must be checked and filled in by people. +- The pre-trained model `facebook/bart-large-cnn` that the app loads was not captured in the BOM by code scanning + alone. It must be collected separately as an inbound material and added (see Considerations below). + +The boundary described in this section, "generation by tools, accuracy and completeness by people," appears +exactly as is in the actual tool output. +{{% /alert %}} + +**Format Sample (CycloneDX ML-BOM)** + +Below is a shortened example of the model component structure of a CycloneDX 1.6 ML-BOM. The key structure +follows the `machine-learning-model` component and `modelCard` in the +[official CycloneDX specification](https://cyclonedx.org/capabilities/mlbom/). If the license is non-standard +(has no SPDX ID), write it as `name`. + +```json +{ + "bomFormat": "CycloneDX", + "specVersion": "1.6", + "components": [ + { + "bom-ref": "model-llama31-8b", + "type": "machine-learning-model", + "group": "meta-llama", + "name": "Llama-3.1-8B", + "version": "1.0", + "licenses": [ + { "license": { "name": "Llama 3.1 Community License" } } + ], + "modelCard": { + "modelParameters": { + "task": "text-generation", + "architectureFamily": "llama", + "datasets": [ + { "type": "dataset", "name": "Public pre-training corpus", "classification": "public" } + ] + }, + "considerations": { + "useCases": ["Summarizing internal documents"], + "technicalLimitations": ["Possibility of hallucination", "Variation in Korean language performance"] + } + } + } + ] +} +``` + +If you use SPDX 3.0, the AI Profile and Dataset Profile express the same information +([SPDX 3.0 AI Profile](https://spdx.github.io/spdx-spec/v3.0.1/model/AI/AI/)). The specific fields of the +formats and the technical details of generation tools are covered in [ISO/IEC 42001 Guide: AI SBOM](../../../iso42001_guide/4-operation/2-ai-sbom/). + +**Considerations** + +- **Account for inbound materials (specification obligation)**: Have a procedure that creates SBOM entries at + the time of intake so that models and datasets brought in from outside are not left out of the AI SBOM. This is a `shall`-level obligation. +- **Generation by tools, review by people**: Generation tools copy the license written in the model card as is. + Because licenses are often missing or incorrectly stated in the model card itself, have people review the + license and provenance fields of the generated AI SBOM against the original sources. +- **Format consistency**: Choose either SPDX or CycloneDX as the default format within the organization, and + operate tools and repositories consistently. Both formats treat models and datasets as first-class components. +- **CI/CD integration**: An AI SBOM is not a deliverable that is created once and finished. Integrate it with + the pipeline so that it is regenerated whenever a model or dataset changes. + +### 3.9.2 Records Demonstrating That the Procedure Was Followed + +**How to Comply** + +Verification material 3.9.2 is the record showing that the procedure was actually followed. Along with the AI +SBOM file itself, keep a history of when and by whom it was generated, reviewed, and approved. If it is generated +automatically in CI/CD, the build logs and the generated SBOM artifacts serve as records, and the history of +uploads to a management tool such as Dependency-Track also serves as evidence. + +**Considerations** + +- **Preserve generation history**: For each version of a supplied AI system, archive the AI SBOM at that point in time to ensure traceability. +- **Approval records**: Record who carried out the review and approval. This connects to the lifecycle review in + [Governance (3.10)](../../4-governance/1-governance/). + +## 5. References + +- License obligation review procedure: [3.5 License obligations](../1-license-obligations/) +- Technical details of AI SBOM formats and generation tools: [ISO/IEC 42001 Guide: AI SBOM](../../../iso42001_guide/4-operation/2-ai-sbom/) +- SBOM management tools: [Dependency-Track](../../../tools/7-dependency-track/), [cdxgen + Dependency-Track integration](../../../tools/8-cdxgen-dt/) +- Governance and lifecycle review: [3.10 Governance](../../4-governance/1-governance/) diff --git a/content/en/guide/ai-sbom_guide/2-ai-extension/_index.md b/content/en/guide/ai-sbom_guide/2-ai-extension/_index.md new file mode 100644 index 0000000000..1da17ba0f5 --- /dev/null +++ b/content/en/guide/ai-sbom_guide/2-ai-extension/_index.md @@ -0,0 +1,15 @@ +--- +title: "AI Extension Processes" +linkTitle: "AI Extension Processes" +weight: 20 +type: docs +categories: ["guide"] +tags: ["AI SBOM", "License", "Transparency"] +description: > + The phase that builds AI-specific license, transparency, and SBOM processes covering models, weights, + and datasets beyond code. +--- + +This is Phase 2 of the implementation roadmap. It is the area where the AI SBOM guide extends ISO/IEC 5230 +the most, and it covers license obligations (3.5), transparency obligations (3.6), and AI SBOM creation +and management (3.9). diff --git a/content/en/guide/ai-sbom_guide/3-relevant-tasks/1-access/_index.md b/content/en/guide/ai-sbom_guide/3-relevant-tasks/1-access/_index.md new file mode 100644 index 0000000000..d50e84f6e3 --- /dev/null +++ b/content/en/guide/ai-sbom_guide/3-relevant-tasks/1-access/_index.md @@ -0,0 +1,123 @@ +--- +title: "3.7 Access" +weight: 10 +type: docs +categories: ["guide"] +tags: ["AI SBOM", "External Inquiry", "Access"] +description: > + Explains how to publicly identify a means for third parties to make AI SBOM compliance inquiries, and how to respond to them internally. +--- + +{{% alert title="Implementation phase" color="info" %}} +This clause is implemented in **Phase 3: Operating Structure**. +[View the full implementation roadmap](../../#phased-implementation-roadmap) +{{% /alert %}} + +## 1. Clause Overview + +Organizations that exchange AI systems in the supply chain need to confirm each other's compliance. For that, a channel for external inquiries must be publicly available, and the organization must be ready to respond to those inquiries. Access ensures both directions. + +3.7 requires two things: publicly identifying a means by which a third party can make an AI SBOM compliance inquiry, and maintaining an internal process to respond effectively to those inquiries. For AI, inquiries also cover AI-specific information such as model and dataset licenses, training data provenance, and model cards. + +## 2. What to Do + +- Publish a means (e.g., a public email address) by which third parties can make AI SBOM compliance inquiries. +- Post the public means where it can be found externally, such as product notices or the website. +- Document the internal procedure for receiving, classifying, and responding to external inquiries. +- Set the response owner and handling deadlines. +- Record inquiries and responses. + +## 3. Requirements and Verification Materials + +| Clause | Requirement | Verification Material(s) | +|-----------|--------------|---------| +| 3.7 | Maintain a process to respond effectively to external AI SBOM compliance inquiries, and publicly identify a means by which third parties can make inquiries. | **3.7.1** Publicly available method that allows interested parties to make an AI SBOM compliance inquiry (e.g., a public contact email)
**3.7.2** Internal documented procedure for responding to third-party AI SBOM compliance inquiries | + +
View original English text + +> **3.7 Access** +> Maintain a process to effectively respond to external AI SBOM Compliance inquiries. Publicly +> identify a means by which a third party can make an AI SBOM Compliance inquiry. +> +> **Verification material(s):** +> - Publicly visible method that allows any interested parties to make an AI SBOM Compliance inquiry +> (e.g., via a published contact email address). +> - An internal documented procedure for responding to third-party AI SBOM Compliance inquiries. + +
+ +## 4. How to Comply and Samples by Verification Material + +### 3.7.1 Public External Inquiry Channel + +**How to Comply** + +Identify a public contact method that anyone can find. A role-based email address (a job function address, not a personal one) is stable. Places to post it include product notices, the open source and AI policy page of the company website, and the contact section of the model card. Stating a response deadline along with it increases trust. + +**Sample** + +``` +AI compliance inquiries: ai-compliance@company.com + +We accept inquiries about the components, model and dataset licenses, and AI SBOMs +of the AI systems we provide. We send an initial reply within 14 business days of receipt. + +(Posted at: product notices, company website AI policy page, model card contact section) +``` + +--- + +### 3.7.2 Internal Inquiry Response Procedure + +**How to Comply** + +Document the internal procedure from receiving an external inquiry to answering it. Define the steps of receipt, classification, assignment, review, reply, and recording, and set a deadline for each step. Because AI SBOM inquiries require referring to model cards and license review records, the AI SBOM verification owner and the license review owner respond together. + +The figure below shows the external inquiry response flow. + +```mermaid +flowchart TD + A([Receive external inquiry\nai-compliance@]) --> B[Classify\nAI SBOM, license, transparency] + B --> C[Assign owner\nverification or legal owner] + C --> D[Review\ncheck AI SBOM and model card] + D --> E{Legal review\nneeded?} + E -->|Yes| F[Legal advice] + E -->|No| G[Draft reply] + F --> G + G --> H([Send and close\nrecord history]) + + style A fill:#2d3748,color:#fff + style D fill:#2b6cb0,color:#fff + style F fill:#744210,color:#fff + style H fill:#22543d,color:#fff +``` + +**Figure 1.** Response flow for external AI SBOM compliance inquiries + +**Considerations** + +- **Set deadlines**: Set a deadline for the initial reply (e.g., 14 days) and for the final answer (e.g., 60 days). +- **Protect AI-specific information**: Set criteria for how much sensitive information, such as model weights or training data, to disclose when answering inquiries. Define the boundary between trade secrets and transparency obligations (3.6). *([Guide recommendation])* +- **Keep records**: Record the inquiry, the reply, and the handling period, and retain them as verification materials. + +**Sample (response procedure outline)** + +``` +## AI SBOM Compliance Inquiry Response Procedure + +1. Receipt: Register inquiries received at ai-compliance@. +2. Classification: Classify as AI SBOM request / license inquiry / transparency obligation inquiry. +3. Assignment: Based on the classification, assign to the AI SBOM verification owner or the license review owner. +4. Review and reply: Check the relevant AI SBOM and model card and reply. Sensitive information follows + the disclosure criteria. Go through legal review if needed. +5. Recording: Retain the inquiry, reply, and handling period. + +Deadlines: initial reply 14 days, final answer 60 days. +``` + +## 5. References + +- Roles and resources of the response owner: [3.8 Effectively Resourced](../2-resourced/) +- AI SBOM used in replies: [3.9 AI SBOM](../../2-ai-extension/3-ai-sbom/) +- Disclosure scope and transparency obligations: [3.6 Transparency Obligations](../../2-ai-extension/2-transparency-obligations/) +- ISO/IEC 5230 example for external inquiries: [ISO/IEC 5230 Conformance Guide: 3.2.1 External Inquiry Response](../../../iso5230_guide/2-relevant-tasks/1-access/) diff --git a/content/en/guide/ai-sbom_guide/3-relevant-tasks/2-resourced/_index.md b/content/en/guide/ai-sbom_guide/3-relevant-tasks/2-resourced/_index.md new file mode 100644 index 0000000000..641e6b6edc --- /dev/null +++ b/content/en/guide/ai-sbom_guide/3-relevant-tasks/2-resourced/_index.md @@ -0,0 +1,160 @@ +--- +title: "3.8 Effectively Resourced" +weight: 20 +type: docs +categories: ["guide"] +tags: ["AI SBOM", "Resources", "Accountability"] +description: > + Explains how to assign accountability, staff, funding, and legal expertise to the AI SBOM compliance program, and how to remediate non-compliance. +--- + +{{% alert title="Implementation phase" color="info" %}} +This clause is implemented in **Phase 3: Operating Structure**. +[View the full implementation roadmap](../../#phased-implementation-roadmap) +{{% /alert %}} + +## 1. Clause Overview + +If competence (3.2) defines "who must be able to do what," effective resourcing attaches actual people, time, and budget to those roles so that the program works. If policies and procedures exist only on paper with no resources, compliance exists in name only. + +3.8 requires assigning accountability for program tasks and allocating sufficient resources. There are five verification materials, covering identification of role owners, provision of staff and funding, access to legal expertise, a procedure for assigning internal responsibilities, and a procedure for remediating non-compliance. The specification cites the resource-related sections (B.4.2, B.4.6) and the human oversight section (B.9.3) of Annex B of ISO/IEC 42001 as references. + +## 2. What to Do + +- Assign accountability to ensure the successful execution of program tasks. +- Allocate sufficient time and funding to the tasks. +- Make legal expertise on AI SBOM compliance accessible to those who need it. +- Have a process for reviewing and updating the policy and tasks. +- Have a process for reviewing and remediating non-compliant cases. + +## 3. Requirements and Verification Materials + +| Clause | Requirement | Verification Material(s) | +|-----------|--------------|---------| +| 3.8 | Assign accountability for program tasks, allocate sufficient time and funding, and provide access to legal expertise and a procedure for remediating non-compliance. | **3.8.1** Document identifying the persons, groups, or functions in program roles
**3.8.2** The identified roles are staffed and adequately funded
**3.8.3** Identification of expertise available to address AI SBOM compliance (internal or external)
**3.8.4** Documented procedure that assigns internal responsibilities for AI SBOM compliance
**3.8.5** Documented procedure for handling the review and remediation of non-compliant cases | + +
View original English text + +> **3.8 Effectively resourced** +> Identify and Resource Program Task(s): assign accountability to ensure the successful execution of +> program tasks; program tasks are sufficiently resourced (time and adequate funding allocated); a +> process exists for reviewing and updating the policy and supporting tasks; legal expertise +> pertaining to AI SBOM Compliance is accessible to those who may need such guidance; and a process +> exists for the resolution of AI SBOM Compliance issues. +> +> **Verification material(s):** +> - Document with name of persons, group or function in program role(s) identified. +> - The identified program roles have been properly staffed and adequate funding provided. +> - Identification of expertise available to address AI SBOM Compliance matters which could be +> internal or external. +> - A documented procedure that assigns internal responsibilities for AI SBOM Compliance. +> - A documented procedure for handling the review and remediation of non-compliant cases. +> +> See, e.g., Sections B.4.2 and B.4.6 of Annex B of ISO/IEC 42001. Section B.9.3 also provides +> guidance to determine if human resources for human oversight should be incorporated. + +
+ +## 4. How to Comply and Samples by Verification Material + +### 3.8.1 Document Identifying Role Owners + +**How to Comply** + +Identify in a document, by name, the persons, groups, or functions that hold program roles. Recording the job title as well is more stable in case people move. In addition to general open source roles, an AI SBOM program needs an AI governance lead, an AI SBOM verification owner, and a model and dataset license review owner. + +**Sample** + +``` +| Role | Owner (job title) | Contact | +|------|-------------|--------| +| AI governance lead | Kim OO (AI Ethics and Governance Lead) | ai-gov@company.com | +| AI SBOM verification owner | Lee OO (Platform Engineer) | sbom@company.com | +| License review owner | Park OO (Open Source Legal) | oss-legal@company.com | +| Security owner | Choi OO (Product Security) | psirt@company.com | +``` + +--- + +### 3.8.2 Staffing and Funding + +**How to Comply** + +Show that the identified roles are actually staffed and budgeted. Record the allocation rate (e.g., 30%) and the basis for the annual budget. Because AI SBOM compliance takes time for model and dataset review and tool operation, estimate the effort for each role realistically. + +**Considerations** + +- **State the allocation rate**: For people with multiple duties, record the share of time allocated to AI SBOM work. +- **Tool budget**: Include the cost of AI SBOM generation and management tools and of legal advice in the budget. + +**Sample** + +``` +| Role | Owner | Allocation | Annual budget basis | Approver/Approval date | +|------|--------|----------|---------------|--------------| +| AI governance lead | Kim OO | 30% | Personnel + regulatory advice | CTO / 2026-01-15 | +| AI SBOM verification owner | Lee OO | 50% | Personnel + tool operation | CTO / 2026-01-15 | +``` + +--- + +### 3.8.3 Access to Legal Expertise + +**How to Comply** + +Make legal expertise on AI SBOM compliance accessible to those who need it. Interpreting non-standard licenses and judging regulatory obligations is the job of legal counsel, so specify the access path, whether internal legal or an external law firm. Also define the escalation criteria (which matters are escalated to legal). + +**Sample** + +``` +- Internal: Open source legal owner (Park OO), first-line license review +- External: AI and IP team of ○○ Law Firm, non-standard license disputes and regulatory interpretation +- Escalation criteria: licenses not on the policy's prohibited or conditional lists, judgments on + downstream disclosure obligations, matters where a regulation newly applies +``` + +--- + +### 3.8.4 Procedure for Assigning Internal Responsibilities + +**How to Comply** + +Document the procedure for assigning internal responsibilities for AI SBOM compliance. Unclear responsibilities lead to gaps, so a RACI matrix that separates Responsible (R), Accountable (A), Consulted (C), and Informed (I) for each task is effective. + +**Sample (RACI matrix)** + +| Task | AI governance lead | AI SBOM verification owner | License review owner | Security owner | +|------|:---:|:---:|:---:|:---:| +| AI SBOM generation | I | R | C | I | +| License obligation review | A | C | R | I | +| Transparency obligation review | A | C | R | I | +| Vulnerability monitoring | I | C | I | R | +| Periodic framework review | R/A | C | C | C | + +*R Responsible, A Accountable, C Consulted, I Informed* + +--- + +### 3.8.5 Procedure for Reviewing and Remediating Non-Compliance + +**How to Comply** + +Have a procedure for reviewing and remediating non-compliant cases (e.g., adopting a model under a prohibited license, a missing AI SBOM, failure to meet transparency obligations). Vary the handling deadline by severity. With AI, license problems sometimes surface after a model has already been deployed, so prepare a remediation path that includes withdrawal or replacement. + +**Sample (remediation procedure and severity criteria)** + +``` +Remediation procedure: identify and report → assess severity → analyze root cause → take corrective action → prevent recurrence → record + +| Severity | Example | Handling deadline | +|--------|------|----------| +| High | A model under a prohibited license is included in an externally distributed product | Immediate response (isolate and review replacement within 48 hours) | +| Medium | Inbound model missing from the AI SBOM | Supplement within 7 days | +| Low | Some model card metadata not filled in | Handle at the next periodic review | +``` + +## 5. References + +- Competence definitions by role: [3.2 Competence](../../1-program-foundation/2-competence/) +- Governance review linked to non-compliance remediation: [3.10 Governance](../../4-governance/1-governance/) +- ISO/IEC 5230 example for resourcing: [ISO/IEC 5230 Conformance Guide: 3.2.2 Effective Resources](../../../iso5230_guide/2-relevant-tasks/2-resourced/) diff --git a/content/en/guide/ai-sbom_guide/3-relevant-tasks/_index.md b/content/en/guide/ai-sbom_guide/3-relevant-tasks/_index.md new file mode 100644 index 0000000000..cdf0ded73f --- /dev/null +++ b/content/en/guide/ai-sbom_guide/3-relevant-tasks/_index.md @@ -0,0 +1,12 @@ +--- +title: "Operations" +linkTitle: "Operations" +weight: 30 +type: docs +categories: ["guide"] +tags: ["AI SBOM", "Operations"] +description: > + This phase sets up a channel for responding to external compliance inquiries and assigns responsibility and resources to the program. +--- + +This is Phase 3 of the implementation roadmap. You set up a channel for responding to external AI SBOM compliance inquiries (3.7) and assign responsibility, staff, and funding to the program (3.8). diff --git a/content/en/guide/ai-sbom_guide/4-governance/1-governance/_index.md b/content/en/guide/ai-sbom_guide/4-governance/1-governance/_index.md new file mode 100644 index 0000000000..1499f26f14 --- /dev/null +++ b/content/en/guide/ai-sbom_guide/4-governance/1-governance/_index.md @@ -0,0 +1,141 @@ +--- +title: "3.10 Governance" +weight: 10 +type: docs +categories: ["guide"] +tags: ["AI SBOM", "Governance", "AI Regulation"] +description: > + Explains how to establish a governance framework across the AI system lifecycle and review it periodically to reflect emerging AI regulations. +--- + +{{% alert title="Implementation phase" color="info" %}} +This clause is implemented in **Phase 4: Governance**. +[View the full implementation roadmap](../../#phased-implementation-roadmap) +{{% /alert %}} + +## 1. Clause Overview + +Governance is the framework that brings all the preceding clauses together and ensures responsible development, deployment, and management across the AI system lifecycle. If the policy (3.1) sets the principles and license obligations (3.5) and the AI SBOM (3.9) create individual processes, governance manages them so that they work consistently even as regulations change and models are replaced. + +3.10 requires having an AI governance framework, policies, and practices. The specification emphasizes compliance with emerging AI laws and regulations such as the EU AI Act, the Hiroshima AI Process, and China's Global AI Governance Initiative, and addresses ethical considerations, risk management, and transparency together. The key point is to review the framework periodically once it is built, so that it reflects the latest regulations and model changes. + +## 2. What to Do + +- Write a governance framework document covering the entire AI system lifecycle. +- Include emerging AI regulation tracking, risk management, transparency, and ethical considerations in the framework. +- Have a procedure for reviewing and updating the framework periodically. +- Monitor the risks associated with ongoing use of AI systems and training data. +- Reflect events such as model tree changes, regulations taking effect, and changes in OSAID classification in governance. *([Guide recommendation])* + +## 3. Requirements and Verification Materials + +| Clause | Requirement | Verification Material(s) | +|-----------|--------------|---------| +| 3.10 | Have an AI governance framework, policies, and practices that ensure AI systems are developed, deployed, and managed responsibly. Emphasize compliance with emerging AI laws (EU AI Act, Hiroshima AI Process, China's initiative), and address ethical considerations, risk management, and transparency. | **3.10.1** Documented AI governance framework for the AI system lifecycle, including a procedure to review the framework periodically | + +
View original English text + +> **3.10 Governance** +> An organization shall have a governance framework for AI, policies, and practices to help ensure +> that AI systems are developed, deployed, and managed responsibly. Governance emphasizes compliance +> with emerging AI laws and regulations, such as the EU AI Act, Hiroshima AI process or Global AI +> Governance Initiative (China), and addresses ethical considerations, risk management, and +> transparency. For example, understand the risks associated with ongoing use of AI Systems and +> training data in the context of their intended Programs. This could include the ability to monitor +> the lifecycle of the AI system and perform ongoing analysis of its intended uses. +> +> **Verification material(s):** +> - A documented AI governance framework for the lifecycle of an AI system with a process to review +> the framework periodically. + +
+ +## 4. How to Comply and Samples by Verification Material + +### 3.10.1 AI Governance Framework and Periodic Review Procedure + +**How to Comply** + +A governance framework contains three parts: tracking emerging regulations to derive obligations, monitoring the AI system lifecycle, and a procedure for periodically reviewing the framework itself. Because regulations change quickly, the framework must not be a fixed document but a system that assumes ongoing updates. + +The three initiatives named in the specification differ in nature. The EU AI Act imposes specific obligations article by article, the Hiroshima AI Process runs voluntary transparency reporting, and China's initiative is closer to a policy declaration. Governance distinguishes these differences and tracks each one accordingly. + +The table below lists the main regulations to track from an AI SBOM perspective. The full regulatory matrix and the ISO/IEC 42001 context are covered in [ISO/IEC 42001 Guide: Context of the Organization and Leadership](../../../iso42001_guide/1-context-leadership/). + +**Table 1.** Main AI regulations that intersect with AI SBOM (as of 2026-06) + +| Regulation or initiative | Timing | Key obligations from an AI SBOM perspective | Reflection in governance | +|---|---|---|---| +| EU AI Act Article 11 + Annex IV | 2027 (high-risk) | Technical documentation obligation | Produce the AI SBOM as a core element of technical documentation | +| EU AI Act Article 53 (GPAI) | 2026-08 | Public summary of training data, respect for copyright opt-outs | Track dataset provenance and licenses | +| EU AI Act Article 50 | 2026-08 | Labeling of AI-generated content | Link with transparency obligations (3.6) | +| Hiroshima AI Process | Launched 2025, Reporting 2.0 (2026-05) | Voluntary transparency reporting | Consider participating in the OECD reporting framework | +| China Global AI Governance Initiative | Announced 2023 | Policy declaration (no specific deliverables) | Monitor developments | +| Korea AI Basic Act (AI 기본법) | In force 2026-01 | Impact assessment for high-impact AI, labeling obligations, disclosure of training data provenance | Prepare AI SBOMs and model cards | + +Lifecycle monitoring means placing governance checkpoints along the flow from development to retirement. The figure below shows lifecycle governance built around the AI SBOM. + +```mermaid +flowchart TD + A([Adopt model or dataset]) --> B[Development: generate AI SBOM\nrecord license and provenance] + B --> C[Review: obligations, risk, regulatory fit\ngovernance lead approval] + C --> D[Deployment: include AI SBOM\nwith supplied system] + D --> E[Operational monitoring\nwatch vulnerabilities and license changes] + E --> F{Change occurred?} + F -->|Model tree change| B + F -->|New regulation in force| G[Update framework] + F -->|No issues| E + G --> H([Periodic review\nquarterly, annual]) + H --> C + + style A fill:#2d3748,color:#fff + style C fill:#744210,color:#fff + style G fill:#c53030,color:#fff + style H fill:#22543d,color:#fff +``` + +**Figure 1.** Lifecycle governance built around the AI SBOM + +**Considerations** + +- **Assign responsibility for regulation tracking**: State in the governance framework who tracks emerging regulations and derives obligations. Obligations under the EU AI Act expand in stages in August 2026 and in 2027, so manage the timing. +- **Manage model tree changes**: When an adopted model moves to a new version or an upstream model is replaced, license obligations may change. Register such changes as governance review events. *([Guide recommendation])* +- **Update OSAID classification**: The distinction between "open source AI" and "open weight" (OSAID 1.0) affects model license judgments. Include changes in classification criteria in the periodic review. +- **State review cycles**: Review model and dataset changes quarterly, and regulations and the framework as a whole annually. Record the review completion date and the reviewer. +- **Link with other clauses**: Governance ties together, at a higher level, the regulatory review in transparency obligations (3.6) and the lifecycle management in AI SBOM (3.9). Link them instead of creating duplicate procedures. + +**Sample (governance framework and annual review plan)** + +Below is a sample of the core parts of a governance framework document and periodic review plan. This document serves as verification material 3.10.1. + +``` +## AI Governance Framework + +### 1. Scope and Purpose +Manage licenses, transparency, risk, and regulatory compliance across the entire lifecycle of AI systems: +adoption, development, deployment, operation, and retirement. + +### 2. Governance Structure +- AI governance lead: approves the framework, makes final judgments on regulatory obligations +- Regulation tracking owner: monitors emerging AI regulations, derives obligations +- AI SBOM verification owner: operates the generation, review, and approval procedure +- Legal: interprets non-standard licenses and regulations + +### 3. Periodic Review Plan +| Cycle | Review items | Owner | Deliverable | +|------|----------|------|--------| +| Quarterly | Model and dataset changes, model tree licenses | AI SBOM verification owner | Change review record | +| Semiannual | Non-standard license classification, OSAID updates | Legal | Updated classification | +| Annual | Regulation enforcement schedule, overall framework, policy consistency | AI governance lead | Revised framework | + +### 4. Change Management +When a model tree change, a new regulation taking effect, or a license policy change occurs, convene an +ad hoc review without waiting for the periodic review. Record review results and actions in the change history. +``` + +## 5. References + +- Policy basis: [3.1 Policy](../../1-program-foundation/1-policy/) +- Transparency obligations and regulatory review: [3.6 Transparency Obligations](../../2-ai-extension/2-transparency-obligations/) +- AI SBOM lifecycle management: [3.9 AI SBOM](../../2-ai-extension/3-ai-sbom/) +- Full regulatory matrix and ISO/IEC 42001 context: [ISO/IEC 42001 Guide: Context of the Organization and Leadership](../../../iso42001_guide/1-context-leadership/) diff --git a/content/en/guide/ai-sbom_guide/4-governance/_index.md b/content/en/guide/ai-sbom_guide/4-governance/_index.md new file mode 100644 index 0000000000..da2b5f3746 --- /dev/null +++ b/content/en/guide/ai-sbom_guide/4-governance/_index.md @@ -0,0 +1,12 @@ +--- +title: "Governance" +linkTitle: "Governance" +weight: 40 +type: docs +categories: ["guide"] +tags: ["AI SBOM", "Governance"] +description: > + This phase establishes a governance framework across the AI system lifecycle and reflects emerging AI regulations. +--- + +This is Phase 4 of the implementation roadmap. You bring the preceding clauses together into a governance framework across the AI system lifecycle, and review it periodically to reflect emerging AI regulations (3.10). diff --git a/content/en/guide/ai-sbom_guide/5-tools/1-aibom-generator/_index.md b/content/en/guide/ai-sbom_guide/5-tools/1-aibom-generator/_index.md new file mode 100644 index 0000000000..55548b5fe8 --- /dev/null +++ b/content/en/guide/ai-sbom_guide/5-tools/1-aibom-generator/_index.md @@ -0,0 +1,96 @@ +--- +title: "OWASP AIBOM Generator" +weight: 10 +type: docs +categories: ["guide"] +tags: ["AI SBOM", "OWASP", "CycloneDX", "Tools"] +description: > + Explains, with screenshots of actual runs, how to use the OWASP tool that generates CycloneDX AI SBOMs from Hugging Face models and scores their completeness. +--- + +## Overview + +OWASP AIBOM Generator is an open source tool that takes a Hugging Face model ID, retrieves the model card metadata, and generates an AI SBOM in CycloneDX format. It is maintained by the OWASP Gen AI Security Project, and its distinguishing feature is scoring how complete the generated BOM is. + +Unlike cdxgen, which identifies dependencies quickly but leaves licenses empty, this tool fills in the license, authors, and external references written in the model card. It is a good starting point for the license review required by [3.5 License Obligations](../../2-ai-extension/1-license-obligations/). + +## Key Features + +- Retrieves metadata of Hugging Face models and generates AIBOMs in CycloneDX 1.6 and 1.7 format. +- Evaluates the completeness of the generated BOM with a score (0 to 100) and a profile, broken down by section. +- Shows model information, the model card, licenses, and external references in a human-readable view. +- Can be used through a web UI or a command line interface (CLI). + +## Usage A: Web UI + +This is the simplest option because you only enter a model ID in the browser. Use the Hugging Face Space provided by the OWASP Gen AI Security Project, or clone the repository and run it locally. + +First, on the input screen, enter a Hugging Face model ID (e.g., `facebook/bart-large-cnn`) and click Generate. + +![OWASP AIBOM Generator input screen](images/input-form.png) + +**Figure 1.** OWASP AIBOM Generator input screen *(GenAI Security Project, captured 2026-06-13)* + +When generation finishes, the result screen shows the AIBOM summary, the completeness assessment, download buttons (CycloneDX 1.6 and 1.7), AI model information, and the model card. The completeness assessment at the top of the screen shows at a glance whether the BOM has the minimum fields needed for identification. + +![OWASP AIBOM Generator result screen](images/result-view.png) + +**Figure 2.** Generation result screen: model information, license (MIT), completeness assessment (Basic) *(captured 2026-06-13)* + +Along with the Human-Friendly View, the result screen provides tabs for a field checklist, a score report, and a JSON view. You can check the fields needed for license obligation review and AI SBOM retention directly on screen and download them as a CycloneDX file. + +## Usage B: Command Line (CLI) + +The CLI is convenient when adding the tool to CI/CD or processing many models in a batch. After installation, pass the model ID as an argument. + +```bash +# Install (a Python virtual environment is recommended) +pip install "git+https://github.com/GenAI-Security-Project/aibom-generator" + +# Generate an AIBOM from a model ID +aibom facebook/bart-large-cnn -o aibom.json +``` + +Below is the actual run result. It generates CycloneDX 1.6 and 1.7, passes schema validation, and shows the completeness score by section. + +```text +$ aibom facebook/bart-large-cnn -o aibom.json + +✅ Successfully generated CycloneDX 1.6 SBOM — Schema Validation (1.6): Valid +✅ Successfully generated CycloneDX 1.7 SBOM — Schema Validation (1.7): Valid + +📊 Completeness Score: 58.7/100 Profile: Basic + - Required Fields: 20/20 + - Metadata: 8/20 + - Component Basic: 17.1/20 + - Component Model Card: 6.7/30 + - External References: 10/10 +``` + +**Figure 3.** CLI run output *(aibom CLI, model facebook/bart-large-cnn, run 2026-06-13)* + +In the generated BOM, the model component has its license and model card filled in. Unlike the cdxgen output, the `licenses` field is not empty. + +```json +{ + "type": "machine-learning-model", + "name": "bart-large-cnn", + "purl": "pkg:huggingface/facebook/bart-large-cnn", + "licenses": [{ "license": { "id": "MIT" } }], + "authors": [{ "name": "facebook" }], + "modelCard": { "modelParameters": { }, "considerations": { } } +} +``` + +## What the Run Results Show + +{{% alert title="The completeness score depends on the quality of the input metadata" color="info" %}} +In the actual run, the completeness score was 58.7/100 (Basic). Required Fields and External References received full marks, but the model card score was low at 6.7/30. This is not a limitation of the tool but the result of the model provider not filling in enough information in the Hugging Face model card. The tool faithfully retrieves the metadata that exists, but it cannot create information that is missing. If the model card is incomplete, a person must check the sources and supplement it. +{{% /alert %}} + +## References + +- AI SBOM generation and management procedure: [3.9 AI SBOM](../../2-ai-extension/3-ai-sbom/) +- License obligation review: [3.5 License Obligations](../../2-ai-extension/1-license-obligations/) +- Another generation tool: [cdxgen](../2-cdxgen/) +- Official: [OWASP AIBOM Generator](https://genai.owasp.org/resource/owasp-aibom-generator/) diff --git a/content/en/guide/ai-sbom_guide/5-tools/2-cdxgen/_index.md b/content/en/guide/ai-sbom_guide/5-tools/2-cdxgen/_index.md new file mode 100644 index 0000000000..ef88a5032b --- /dev/null +++ b/content/en/guide/ai-sbom_guide/5-tools/2-cdxgen/_index.md @@ -0,0 +1,90 @@ +--- +title: "cdxgen" +weight: 20 +type: docs +categories: ["guide"] +tags: ["AI SBOM", "cdxgen", "CycloneDX", "Tools"] +description: > + Explains, with actual run output, how to use the AI BOM mode of OWASP cdxgen, which generates CycloneDX SBOMs from projects and models. +--- + +## Overview + +cdxgen is the official SBOM generator of the OWASP CycloneDX project. It supports more than 20 languages and package managers, and recent versions provide a dedicated AI BOM mode. It automatically identifies the dependencies of AI applications (PyTorch, Transformers, etc.) and integrates well with CI/CD pipelines. + +From an AI SBOM perspective, cdxgen's strengths are speed and automation. Its weakness is that it does not fill in license information in a default run. The run results below show this characteristic. Whereas OWASP AIBOM Generator centers on model card metadata, cdxgen centers on code and dependencies. Using both together covers both models and dependencies. + +## Key Features + +- Identifies dependencies in source code and container images and generates CycloneDX SBOMs. +- Includes AI/ML metadata (formulation) in AI BOM mode (`-t ai`). +- Accepts Hugging Face model URLs, Modelfiles, and GGUF artifacts directly as input. +- Can automatically submit SBOMs to a Dependency-Track server for ongoing management. + +## Installation + +```bash +# One-time run (requires Node.js) +npx @cyclonedx/cdxgen@latest --version + +# Global installation +npm install -g @cyclonedx/cdxgen +``` + +## Usage: Generating an AI BOM + +Run it in AI BOM mode in the AI project directory. + +```bash +# Generate an AI BOM +cdxgen -t ai -o aibom.json . + +# Generate including AI/ML metadata (formulation) +cdxgen -t ai --include-formulation -o aibom.json . +``` + +Below is the result of actually running cdxgen on a summarization app (depending on `transformers` and `torch`) that loads a pretrained model (`facebook/bart-large-cnn`). It automatically identifies 5 dependencies and creates a CycloneDX 1.7 BOM. + +```text +$ cdxgen -t python --include-formulation -o aibom.json . +CycloneDX Generator 12.5.1 (Node.js) + +Generated components: 5 (CycloneDX 1.7): + transformers 4.44.2 pkg:pypi/transformers@4.44.2 license: empty + torch 2.4.0 pkg:pypi/torch@2.4.0 license: empty + numpy 1.26.4 pkg:pypi/numpy@1.26.4 license: empty + tokenizers 0.19.1 pkg:pypi/tokenizers@0.19.1 license: empty + huggingface-hub 0.24.6 pkg:pypi/huggingface-hub@0.24.6 license: empty +``` + +**Figure 1.** cdxgen run output *(cdxgen 12.5.1, run 2026-06-13)* + +One of the generated components looks like this. The identification evidence (evidence) is filled in, but the `licenses` field is empty. + +```json +{ + "name": "transformers", + "version": "4.44.2", + "purl": "pkg:pypi/transformers@4.44.2", + "type": "library", + "evidence": { + "identity": [ + { "field": "purl", "confidence": 0.5, + "methods": [{ "technique": "manifest-analysis", "value": "requirements.txt" }] } + ] + } +} +``` + +## What the Run Results Show + +{{% alert title="cdxgen identifies quickly, but a person must supplement the licenses" color="warning" %}} +cdxgen quickly identified 5 dependencies from `requirements.txt`, but the `licenses` field of each component is empty. In addition, the pretrained model `facebook/bart-large-cnn` that the app loads was not captured in the BOM by code scanning alone. It must be collected separately as an inbound component and added. A practical combination is to build the dependency skeleton quickly with cdxgen, have a person check and fill in the licenses, and generate the model separately with OWASP AIBOM Generator and merge it. +{{% /alert %}} + +## References + +- AI SBOM generation and management procedure: [3.9 AI SBOM](../../2-ai-extension/3-ai-sbom/) +- Generator centered on model metadata: [OWASP AIBOM Generator](../1-aibom-generator/) +- SBOM management: [Dependency-Track](../../../tools/7-dependency-track/), [cdxgen + Dependency-Track integration](../../../tools/8-cdxgen-dt/) +- Official: [cdxgen AI-BOM documentation](https://github.com/cdxgen/cdxgen/blob/master/docs/AI_BOM.md) diff --git a/content/en/guide/ai-sbom_guide/5-tools/3-scanners/_index.md b/content/en/guide/ai-sbom_guide/5-tools/3-scanners/_index.md new file mode 100644 index 0000000000..245561f210 --- /dev/null +++ b/content/en/guide/ai-sbom_guide/5-tools/3-scanners/_index.md @@ -0,0 +1,72 @@ +--- +title: "Model and Container Scanners (Lab700x, Trivy, Syft)" +linkTitle: "Model and Container Scanners" +weight: 30 +type: docs +categories: ["guide"] +tags: ["AI SBOM", "Security", "Tools"] +description: > + Introduces the main features and usage of security scanners that analyze AI model binaries, inference servers, and AI packages. +--- + +This page introduces analysis and identification tools that complement AI SBOM generation. If the generation tools (OWASP AIBOM Generator, cdxgen) record "what is inside," these tools look at "whether it is safe" and "which version it is." The commands and features below are based on each tool's official documentation (the tools actually run in this guide are [OWASP AIBOM Generator](../1-aibom-generator/) and [cdxgen](../2-cdxgen/)). + +## Lab700x AI SBOM Scanner: Static Analysis of Model Binaries + +This tool statically analyzes AI model files themselves to extract information. It examines model binaries such as `.safetensors`, `.pt` (PyTorch), and `.pkl` (Pickle) directly without running them (deep introspection). + +- **Key features**: Because it understands the internal structure without executing the model, it detects malicious code hidden in model files (such as Pickle injection), vulnerabilities, and license violations before deployment. +- **Role in AI SBOM**: Used to inspect externally adopted models at the intake gate. Combined with the enforcement of intake metadata in [3.5 License Obligations](../../2-ai-extension/1-license-obligations/), it lets you perform metadata verification and binary safety inspection together. + +Models in Pickle format can execute arbitrary code during deserialization, so the supply chain risk is high. The core of this tool is that it inspects models without running them. + +## Trivy: Scanning LLM Inference Server Containers + +An open source scanner from Aqua Security that checks container images and file systems for vulnerabilities. It recently added awareness of AI model infrastructure. + +- **Key features**: Scans LLM inference server containers such as Ollama and LocalAI and collects the open source libraries they contain and container vulnerabilities. +- **Usage**: + +```bash +# Scan a container image (vulnerabilities) +trivy image ollama/ollama:latest + +# Generate an SBOM (CycloneDX) +trivy image --format cyclonedx --output sbom.json ollama/ollama:latest +``` + +In environments that deploy AI models as containers, it is used to record the vulnerabilities and components of the inference server as an SBOM. + +## Syft: Identifying AI Packages and Virtual Environments + +Anchore's SBOM generator, which scans containers, file systems, and virtual environments. + +- **Key features**: Analyzes Python virtual environments and collects the exact versions of packages used to run AI, such as PyTorch and Transformers. Outputs in SPDX and CycloneDX formats. +- **Usage**: + +```bash +# Scan a directory and output CycloneDX +syft scan dir:. -o cyclonedx-json=sbom.json + +# Scan a container image +syft scan registry:python:3.11-slim -o spdx-json +``` + +Its role is similar to cdxgen, so for organizations that already use the Anchore toolset (Syft, Grype), a natural combination is to generate the dependency SBOM of AI applications with Syft and check vulnerabilities with Grype. + +## Recommended Tool Combination + +No single tool covers everything. In practice, divide the roles and combine tools. + +| Purpose | Tool | +|------|------| +| AIBOM generation from model metadata | OWASP AIBOM Generator | +| Dependency SBOM generation | cdxgen, Syft | +| Model binary safety inspection | Lab700x AI SBOM Scanner | +| Inference server and container vulnerabilities | Trivy | +| SBOM storage and vulnerability monitoring | [Dependency-Track](../../../tools/7-dependency-track/) | + +## References + +- AI SBOM generation and management procedure: [3.9 AI SBOM](../../2-ai-extension/3-ai-sbom/) +- Run results of generation tools: [OWASP AIBOM Generator](../1-aibom-generator/), [cdxgen](../2-cdxgen/) diff --git a/content/en/guide/ai-sbom_guide/5-tools/_index.md b/content/en/guide/ai-sbom_guide/5-tools/_index.md new file mode 100644 index 0000000000..25e1cf56d7 --- /dev/null +++ b/content/en/guide/ai-sbom_guide/5-tools/_index.md @@ -0,0 +1,24 @@ +--- +title: "Tools" +linkTitle: "Tools" +weight: 50 +type: docs +categories: ["guide"] +tags: ["AI SBOM", "Tools"] +description: > + Explains the main features and usage of open source tools that generate and analyze AI SBOMs, with screenshots of actual runs. +--- + +This section covers open source tools that automate AI SBOM compliance. For each tool, it summarizes the main features, installation, and usage, along with actual run results. It makes the categories in the automation maturity map of [3.9 AI SBOM](../2-ai-extension/3-ai-sbom/) concrete, tool by tool. + +There is a limit worth stating plainly. Tools generate BOMs automatically, but they cannot guarantee that the licenses in the generated BOM are correct or that no components are missing. In the tool comparison below, OWASP AIBOM Generator fills in the license from the model card, while cdxgen identifies dependencies quickly but leaves the license field empty. Consider this difference when choosing a tool. + +## Tools at a Glance + +| Tool | Input | Output | Strengths | Covered in | +|---|---|---|---|---| +| OWASP AIBOM Generator | Hugging Face model ID | CycloneDX 1.6/1.7 | Model card and license metadata, completeness score | [Go](./1-aibom-generator/) | +| cdxgen | Project directory, model files | CycloneDX | Automatic dependency identification, CI/CD integration | [Go](./2-cdxgen/) | +| Lab700x, Trivy, Syft | Model binaries, containers, virtual environments | Reports, SBOMs | Static analysis of models, identification of inference servers and packages | [Go](./3-scanners/) | + +Each tool automates some of the generation, analysis, and management stages. No single tool covers everything, so combine tools that generate AI SBOMs (OWASP AIBOM Generator, cdxgen), tools that analyze security (Lab700x, Trivy), and tools that manage them ([Dependency-Track](../../tools/7-dependency-track/)). diff --git a/content/en/guide/ai-sbom_guide/_index.md b/content/en/guide/ai-sbom_guide/_index.md new file mode 100644 index 0000000000..b8caa7c3ae --- /dev/null +++ b/content/en/guide/ai-sbom_guide/_index.md @@ -0,0 +1,187 @@ +--- +title: "AI SBOM Compliance Guide" +linkTitle: "AI SBOM Compliance Guide" +weight: 43 +type: docs +categories: ["guide"] +tags: ["AI SBOM", "OpenChain", "ISO/IEC 5230", "ISO/IEC 42001", "Compliance"] +description: > + A practical guide for organizations that explains, clause by clause, the requirements of the + OpenChain AI SBOM Compliance Guide (Version 1.0). +--- + +This guide explains, one by one, each requirement of *AI System Bill of Materials: Compliance +Management Guide for the Supply Chain* (Version 1.0), published by the OpenChain AI Work Group. +It describes what verification materials each clause requires, how to comply, and what samples and +tools can be used directly. + +This specification carries the structure of the open source license compliance standard ISO/IEC 5230 +(requirements, verification materials, rationale) over to the AI supply chain. It brings into the scope +of compliance not only code but also the license and transparency obligations of model weights, +training datasets, and the Model Tree. + +**Author : OpenChain Korea Work Group / [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)** + +{{% alert title="Note" color="info" %}} +All 10 requirements (3.1 to 3.10), numbered as in the body of the specification, have been written. +The compare page that compares the positioning of the standards is being expanded. +{{% /alert %}} + +## Target Audience + +- Compliance practitioners at organizations that develop AI systems or exchange them through the supply chain +- Practitioners who have an open source compliance (ISO/IEC 5230) program in place and want to extend it to AI +- Legal, security, and development staff who need to review the license and transparency obligations of AI models and datasets + +## How to Use This Guide + +{{% alert title="Division of roles between the OpenChain specification and the KWG practical guide" color="success" %}} +The OpenChain specification defines "what must be demonstrated." This guide fills in "how to achieve +it." Each clause page does not stop at restating the specification requirements. It also covers actual +procedures, samples, tools, and how to handle the parts that tools alone cannot cover. +{{% /alert %}} + +{{% alert title="Notation: [Specification requirement] vs. [Guide recommendation]" color="info" %}} +The content of each clause page is divided into the following two categories. + +- **[Specification requirement]**: Items that the body of the AI SBOM Compliance Guide states with `shall` or as verification materials. +- **[Guide recommendation]**: Items that are not in the body of the specification but that the OpenChain Korea Work Group + recommends based on practical experience, good practices, and other standards (ISO/IEC 5230, 42001, etc.). Adoption is at the organization's discretion. + +Activities presented with a verification material number (e.g., `3.1.1`) are [Specification requirement]. Additions +made by this guide, such as automation, tool use, and intake gates, are [Guide recommendation]. +{{% /alert %}} + +{{% alert title="About clause numbering" color="info" %}} +In the original specification, the section numbers in the table of contents and the body do not match (the +"3.9 AI content review and approval" section in the table of contents is missing from the body, so the +following numbers shift up by one). This inconsistency has been reported to the OpenChain AI Work Group. +This guide follows the numbering of the specification body (3.1 to 3.10). +{{% /alert %}} + +## Phased Implementation Roadmap + +The 10 requirements are divided into four phases by implementation priority. Phase 1 lays the foundation +of the program, Phase 2 builds the AI-specific compliance processes, Phase 3 establishes the operating +structure, and Phase 4 puts governance in place. + +--- + +### Phase 1: Program Foundation + +**Goal**: Define the scope of the program, establish the policy, and secure competence and awareness. + +| Done | Verification Material | Description | Detailed Guide | +|:----:|---------|------|------------| +| ☐ | **3.4.1** | Program scope statement | [3.4 →](./1-program-foundation/4-scope/) | +| ☐ | **3.1.1** | Documented AI SBOM policy | [3.1 →](./1-program-foundation/1-policy/) | +| ☐ | **3.1.2** | Policy awareness procedure | [3.1 →](./1-program-foundation/1-policy/) | +| ☐ | **3.2.1-3.2.3** | List of roles, competence definitions, evidence of assessed competence | [3.2 →](./1-program-foundation/2-competence/) | +| ☐ | **3.3.1** | Evidence of assessed participant awareness | [3.3 →](./1-program-foundation/3-awareness/) | + +--- + +### Phase 2: AI Extension Processes + +**Goal**: Build AI-specific license, transparency, and SBOM processes that cover models, weights, and +datasets beyond code. This is the area where the AI SBOM guide extends ISO/IEC 5230 the most. + +| Done | Verification Material | Description | Detailed Guide | +|:----:|---------|------|------------| +| ☐ | **3.5.1** | Procedure for reviewing and documenting license obligations | [3.5 →](./2-ai-extension/1-license-obligations/) | +| ☐ | **3.6.1** | Procedure for reviewing transparency obligations | [3.6 →](./2-ai-extension/2-transparency-obligations/) | +| ☐ | **3.9.1** | Procedure for identifying, tracking, reviewing, approving, and archiving the AI SBOM | [3.9 →](./2-ai-extension/3-ai-sbom/) | +| ☐ | **3.9.2** | Records demonstrating that the procedure was followed | [3.9 →](./2-ai-extension/3-ai-sbom/) | + +--- + +### Phase 3: Operating Structure + +**Goal**: Set up a channel for responding to external compliance inquiries, and assign accountability and resources to the program. + +| Done | Verification Material | Description | Detailed Guide | +|:----:|---------|------|------------| +| ☐ | **3.7.1-3.7.2** | Public inquiry method, internal response procedure | [3.7 →](./3-relevant-tasks/1-access/) | +| ☐ | **3.8.1-3.8.5** | Role assignment, resources, legal expertise, remediation procedure | [3.8 →](./3-relevant-tasks/2-resourced/) | + +--- + +### Phase 4: Governance + +**Goal**: Establish a governance framework across the AI system lifecycle and reflect emerging AI regulations. + +| Done | Verification Material | Description | Detailed Guide | +|:----:|---------|------|------------| +| ☐ | **3.10.1** | AI governance framework and periodic review procedure | [3.10 →](./4-governance/1-governance/) | + +--- + +## Full Clause Checklist + +The body of the AI SBOM Compliance Guide consists of **10 clauses and 19 verification material items** (based on +the verification material numbering used in this guide). + +| Clause | Title | Verification Materials | Details | +|------|------|:---:|------| +| 3.1 | Policy | 2 | [Go to](./1-program-foundation/1-policy/) | +| 3.2 | Competence | 3 | [Go to](./1-program-foundation/2-competence/) | +| 3.3 | Awareness | 1 | [Go to](./1-program-foundation/3-awareness/) | +| 3.4 | Program scope | 1 | [Go to](./1-program-foundation/4-scope/) | +| 3.5 | License obligations | 1 | [Go to](./2-ai-extension/1-license-obligations/) | +| 3.6 | Transparency obligations | 1 | [Go to](./2-ai-extension/2-transparency-obligations/) | +| 3.7 | Access | 2 | [Go to](./3-relevant-tasks/1-access/) | +| 3.8 | Effectively resourced | 5 | [Go to](./3-relevant-tasks/2-resourced/) | +| 3.9 | AI SBOM | 2 | [Go to](./2-ai-extension/3-ai-sbom/) | +| 3.10 | Governance | 1 | [Go to](./4-governance/1-governance/) | + +**Total: 10 clauses / 19 verification material items** + +## Automation Maturity Map + +This map plainly distinguishes how far each AI SBOM compliance task can be automated with tools. +"Generation" already has usable open source tools. In contrast, interpreting license obligations and +tracking compliance with non-standard licenses are still the job of people and policy. Each clause page +follows this boundary and separates "what tools can do" from "what people must fill in." + +| Task | Automation Level | Representative Open Source Tools | +|------|------------|-------------------| +| Generating code and dependency SBOMs | Mature | cdxgen, Syft | +| Generating BOMs for AI models and metadata | Tools emerging | OWASP AIBOM Generator, cdxgen `aibom` mode | +| Static analysis of model binaries | Tools emerging | Lab700x AI SBOM Scanner | +| Identifying LLM inference servers and AI packages | Mature | Trivy, Syft | +| SBOM storage and vulnerability monitoring | Mature | Dependency-Track, SW360 | +| Interpreting license obligations and tracking non-standard compliance | Immature (people and policy) | Tools at a supplementary stage | + +{{% alert title="Generation by tools, interpretation by people" color="info" %}} +Several tools already generate AI SBOMs automatically. However, tools cannot automatically guarantee +whether the license fields of a generated BOM are accurate, whether the behavioral use restrictions of +non-standard licenses (the RAIL family, the Llama Community License) are being followed, or whether +obligations are passed downstream without omission. This area is covered by policy and human review. +Details are covered in [3.5 License obligations](./2-ai-extension/1-license-obligations/). +{{% /alert %}} + +Installation and usage of each tool are covered with screenshots and command output in the [Tools](./5-tools/) section. + +## Relationship with Other Standards + +{{% alert title="Relationship with ISO/IEC 5230 and 42001" color="info" %}} +- **ISO/IEC 5230 (license compliance)**: The AI SBOM guide inherits the 5230 methodology as is. + Organizations that already have a 5230 program can reuse the program foundation, such as policy, + competence, and resources, and only need to add the AI extension areas. See the + [ISO/IEC 5230 Conformance Guide](../iso5230_guide/). +- **ISO/IEC 42001 (AI management system)**: Technical details of AI SBOM formats (SPDX 3.0 AI Profile, + CycloneDX ML-BOM) and generation tools are covered in + [ISO/IEC 42001 Guide: AI SBOM](../iso42001_guide/4-operation/2-ai-sbom/). Building on that, this + guide focuses on "how to operate a compliance program." +{{% /alert %}} + +## Original Specification + +{{% pageinfo %}} +- **Document**: Artificial Intelligence System Bill of Materials: Compliance Management Guide for + the Supply Chain, Version 1.0 +- **Publisher**: OpenChain Project AI Work Group, 2025-10-20 +- **License**: Creative Commons Attribution 4.0 (CC-BY-4.0) +- **Official version**: Published as PDF and Markdown in the OpenChain Reference-Material repository (`AI-SBOM-Compliance/en`) +- **Announcement**: [openchainproject.org](https://openchainproject.org/news/2025/10/20/welcoming-the-openchain-ai-system-bill-of-materials-compliance-guide) +{{% /pageinfo %}} diff --git a/content/en/guide/finance-oss-guide/0-closed-network/_index.md b/content/en/guide/finance-oss-guide/0-closed-network/_index.md new file mode 100644 index 0000000000..f6741d67fe --- /dev/null +++ b/content/en/guide/finance-oss-guide/0-closed-network/_index.md @@ -0,0 +1,343 @@ +--- +title: "Closed Network Operation and the Network Separation Transition" +linkTitle: "0. Closed Network Operation" +weight: 10 +type: docs +categories: ["guide"] +tags: ["Finance", "Open Source", "Network Separation", "Closed Network", "Supply Chain Security"] +description: > + Covers the common infrastructure for importing, mirroring, and checking open source in financial sector closed networks, and how to + document a self risk assessment in a network separation regulatory environment that is shifting toward voluntary security. +--- + +{{% alert title="Where this page fits" color="info" %}} +This guide covers the use and management of open source in the financial sector. Its starting point is the closed network. +Import, internal mirrors, and offline vulnerability management are prerequisites for every stage, including identification, use approval, and management, +so they are explained once here, and each stage page points to this page with an "Applying in a closed network" box. + +Documents produced here: an import procedure and import approval records, and a self risk assessment for a network separation exception. +{{% /alert %}} + +## Why cover closed networks first? + +Typical open source practice assumes an internet connection. Developers download dependencies with +package managers, query online vulnerability databases, and use cloud-based scanners. In the critical +business networks and core systems of financial institutions, this assumption does not hold. In an +environment where external communication is blocked, bringing in open source is itself subject to +control, and how to obtain vulnerability information becomes the first problem to solve. + +The "Guide to Using and Managing Open Source Software in the Financial Sector" (금융분야 오픈소스 +소프트웨어 활용·관리 안내서; Financial Supervisory Service (금융감독원) and Financial Security Institute +(금융보안원), 2022, hereafter the FSEC guide) sets out the management procedure +in four stages: identification, issue identification and resolution, use approval, and management. In a +closed network, however, there is a gate to pass before "identification": how the software will be +brought in. That is why this guide moves the closed network up to section 0. + +{{% alert title="Nature of this guide" color="warning" %}} +This guide is supplementary material that does not replace the FSEC guide, and its recommendations are +good practice, not legal obligations (see [Nature and scope in the overview](../)). Most of the closed +network procedures on this page are [Guide recommendation] items that the standards do +not directly require, and only the points that the standards or regulations directly require are marked +with their source as **[ISO requirement]** or [FSEC guide]. +{{% /alert %}} + +## Two realities to handle together + +If you assume closed networks mean only "fixed physical network separation," the guide will soon be out +of date, because the regulatory environment is in transition. The Financial Services Commission +(금융위원회) announced the "Roadmap for Improving Network Separation in the Financial Sector" (금융분야 +망분리 개선 로드맵) (2024-08-13), and amendments to the Regulation on Supervision of Electronic Financial +Transactions (전자금융감독규정) and its Enforcement Rules (시행세칙) then took effect on 2025-02-05. With +these amendments, regulation that had long centered on physical network separation began to shift +toward voluntary security and a risk-based approach. For research and development work that does not +process unique identification information or personal credit information, a financial institution can +apply a network separation exception after a self risk assessment, application of information +protection controls that substitute for network separation, and approval by the Information Protection +Committee (정보보호위원회). Other relaxations, such as expanding the use of cloud-based application +software (SaaS, Software as a Service) on internal business networks, are being pursued as later stages +of the roadmap. + +Practitioners in the financial sector therefore work with two realities at the same time. + +First, closed networks remain in place for critical business and core systems. Import control, internal +mirrors, and offline vulnerability management are still needed. + +Second, in areas where an exception has been opened, such as research and development, it is now +possible to work without network separation. However, relaxed regulation does not mean exemption from +controls. For the parts that network separation used to protect automatically, the organization must now +assess the risk itself, design controls, and prove that they are implemented. Responsibility has moved +from following rules to proving compliance on its own. + +## Import control + +In a closed network, open source is brought in by receiving it in the external zone, verifying and +isolating it, and then transferring it to the internal network through a network linkage system (data +transfer between networks). + +```mermaid +flowchart LR + subgraph EXT["External zone (internet connected)"] + A["Receive from official distribution site"] --> B["Integrity verification
(checksum comparison)"] + B --> C["Generate SBOM
Pre-check vulnerabilities"] + end + C --> D["Assemble import bundle
Malware scan"] + D --> E["Data transfer between networks"] + subgraph INT["Internal network"] + F["Recheck hashes
Review results"] --> G["Register in internal mirror"] + G --> H["Import approval record
(audit evidence)"] + end + E --> F + + style EXT fill:#f7fafc,stroke:#a0aec0 + style INT fill:#f7fafc,stroke:#a0aec0 + style D fill:#744210,color:#fff + style E fill:#2d3748,color:#fff + style H fill:#276749,color:#fff +``` + +The import procedure includes the following. + +- Integrity verification: Compare the hash of the downloaded artifact with the value published by the official distribution site. +- Malware scan: Run antivirus and malware scans in the import zone. +- Component identification: Record the components of the imported artifact as an SBOM (Software Bill of Materials). + This SBOM becomes a verification material required by SBOM management in ISO/IEC 5230 (§3.3.1). **[ISO requirement]** +- Import approval record: Record who imported what and why, together with the verification results. This record becomes audit evidence. + +### Example import procedure + +As an example, consider moving one library from the external zone to the internal network. Tool names +are examples, and you can replace them with equivalent tools your organization already uses. + +In the external zone (the area with internet access), receive the artifact and check its integrity. + +```bash +# 1) Download the artifact together with its signature and checksum from the official distribution site +curl -LO https://example.org/lib/foo-1.2.3.tar.gz +curl -LO https://example.org/lib/foo-1.2.3.tar.gz.sha256 + +# 2) Verify integrity by comparing with the published hash +sha256sum -c foo-1.2.3.tar.gz.sha256 + +# 3) Record the components of the import unit as an SBOM (CycloneDX format) +syft foo-1.2.3.tar.gz -o cyclonedx-json > foo-1.2.3.sbom.json + +# 4) Pre-check vulnerabilities in the external zone and include the results in the import bundle +grype sbom:foo-1.2.3.sbom.json -o json > foo-1.2.3.vuln.json +``` + +The SBOM at this stage is a record that identifies the imported artifact itself. A single source +bundle does not include the other components the library depends on, so transitive dependencies are +not expanded by this command. Each transitive dependency comes in as a separate import unit and is +registered in the internal mirror, and the dependencies of an entire project are identified by +generating an SBOM for the project at build time (section 2). + +Package the artifact, SBOM, vulnerability check results, and checksum into one import bundle, run a +malware scan on it, and move it to the internal network through data transfer between networks. On the +internal network, recheck the hashes, review the SBOM and vulnerability results, and then register the +bundle in the internal mirror repository. Record the import approval at the same time. + +{{% alert title="What a new organization should do first, and how an operating organization can advance" color="success" %}} +An organization setting up a system for the first time starts by consolidating import into a single +path and defining a standard form for what an import bundle contains (artifact, hash, SBOM, +vulnerability results, approval record). + +An organization that already operates a system automates the import procedure, enforces that only +pre-approved components are supplied through the mirror, and links import records to its audit evidence +system. +{{% /alert %}} + +## Internal mirror repository + +Set up an internal mirror repository that supplies open source inside the closed network. Run a +repository manager such as Nexus or Artifactory on premises, or mirror language-specific package +repositories internally. Developers obtain dependencies only from this internal mirror, not from the +internet. + +The mirror is both a repository that stores open source and a control point for what is used and in +which version. + +- Register the SBOM created at the import stage in the mirror together with the artifact, to fix which versions are present. +- Identify transitive dependencies as well (lower-level dependencies that come in without being declared directly). +- Do not put components that have not been pre-approved into the mirror, to prevent use that bypasses the approval procedure. + +With an internal mirror, what developers use can be seen in one place, and it becomes the basis for +identification (section 2) and use approval (section 4). + +## Offline vulnerability management + +A closed network cannot receive vulnerability information in real time. Because there is a structural +limit in that new vulnerabilities (CVE, Common Vulnerabilities and Exposures) are recognized later, set +up a separate procedure for bringing in vulnerability data regularly. + +There are two approaches. + +- Import only the results of scans run on the external network. In an area with internet access, check + the SBOM for vulnerabilities and move the resulting report to the internal network. +- Synchronize an offline vulnerability database regularly. Download the database that the vulnerability + checking tool references from outside, import it into an internal mirror, and point internal tools to that mirror. + +[Grype](https://github.com/anchore/grype) and [Trivy](https://trivy.dev/) support offline updates in +physically separated closed network (air-gap) environments by downloading the vulnerability +database as a cache directory and moving it. +[Dependency-Track](../../tools/7-dependency-track/) can point to internal mirrors by changing, in its +configuration, the addresses of the vulnerability data sources it references (the National Vulnerability +Database, OSV, GitHub Advisories, and others), but this is harder because a mirror must be set up +separately for each data source. A small-scale configuration test before adoption is recommended. +Whether a tool supports offline updates is central to vulnerability management in a closed network. + +### Example of importing an offline vulnerability database + +As an example, consider downloading only the vulnerability database from outside and moving it inside. + +Download the database in the external zone. + +```bash +# External zone: download the vulnerability DB archive (tool-specific commands are examples) +grype db update # Download the latest vulnerability DB to the local cache +grype db status # Check the version and build time of the downloaded DB + +# Package the downloaded DB cache directory into an archive to be imported +tar czf grype-db-$(date +%Y%m%d).tar.gz -C ~/.cache/grype/db . +``` + +Check the hash of the archive, run a malware scan, and move it to the internal network. Then configure +the internal checking tool to reference this database. + +```bash +# Internal network: extract the imported DB archive to the location the checking tool references +tar xzf grype-db-YYYYMMDD.tar.gz -C /opt/grype/db + +# Check in offline mode (specify the location of the imported DB and turn off network updates) +GRYPE_DB_CACHE_DIR=/opt/grype/db GRYPE_DB_AUTO_UPDATE=false grype sbom:foo-1.2.3.sbom.json +``` + +For Trivy, only the command for downloading the database differs, and the flow is the same. Download +and package the cache in the external zone, then extract it inside, specify the cache location, and run +the check. + +```bash +# External zone: download the Trivy vulnerability DB to the local cache and package it into an archive +trivy image --download-db-only +tar czf trivy-db-$(date +%Y%m%d).tar.gz -C ~/.cache/trivy . + +# Internal network: extract the cache and check with the imported DB without updating +tar xzf trivy-db-YYYYMMDD.tar.gz -C /opt/trivy +trivy sbom foo-1.2.3.sbom.json --skip-db-update --cache-dir /opt/trivy +``` + +Checking Java projects requires a separate Java index database, so download it as well in the external +zone with `trivy image --download-java-db-only` and include it in the same cache. + +Define the database synchronization interval and the person responsible, and manage the delay in +awareness that can occur while synchronization lags. If the SBOMs of production systems are registered +in Dependency-Track, each database update automatically reassesses new vulnerabilities that affect +systems already in operation. Continuous monitoring is covered in detail in section 5. + +## Choosing tools for a closed network + +Cloud-based scanners cannot be used in a closed network, so use tools that are installed on premises. If +a tool is itself open source, the tool also becomes subject to import. + +Most of the tools this guide covers on its tool pages are open source and can be installed on premises. +For SBOM generation you can use [Syft](../../tools/6-syft/), [cdxgen](../../tools/5-cdxgen/), and +[OSV-SCALIBR](../../tools/4-osvscalibr/); for license checks, +[FOSSology](../../tools/1-fossology/) and [SCANOSS](../../tools/9-scanoss/); and for vulnerability +checks and continuous monitoring, [Dependency-Track](../../tools/7-dependency-track/), +[Grype](https://github.com/anchore/grype), and [Trivy](https://trivy.dev/). Installation and usage for +each tool are covered at each link and on the [tool pages](../../tools/). + +A dedicated platform is not the only answer. If you run a CI platform such as GitLab in house, you can +set up continuous checks with its built-in dependency scanning feature, and commercial SCA tools play the +same role. Such combinations are also common in the tool setups of actual financial institutions (see +[field cases in Management](../5-manage/)). + +We do not recommend any single specific product. When choosing tools for a closed network, look at the +following criteria. + +- Is it provided as an on-premises installation? +- Can its vulnerability database be updated offline? +- Does it import and export standard SBOM formats (SPDX, CycloneDX)? +- Does the tool's own license cause problems for in-house operation? + +The last criterion does come up in practice. For example, [FOSSLight](../../tools/3-fosslight/) is +AGPL-3.0. If a modified version of the tool is used to provide functionality over a network, a source +disclosure obligation can arise, so treat it as an item for legal review. + +{{% alert title="Reviewing commercial tools" color="info" %}} +Financial institutions often also review commercial software composition analysis (SCA, Software +Composition Analysis) tools because of service level agreements (SLA, Service Level Agreement), Korean +language support, and accountability. This guide does not recommend specific commercial products. In +addition to the selection criteria above, it advises weighing whether closed network installation is +provided, offline database updates, local support, and support for standard output. It treats an open +source stack as the default recommendation and commercial tools as an option to supplement it according +to the organization's needs. +{{% /alert %}} + +## Managing patch delays + +In a closed network, vulnerability patches must also go through the import procedure again, so an +immediate response is difficult. Even after a new vulnerability is disclosed, it takes time to obtain +the patch, verify it, and move it inside. Put measures in place to manage this delay. + +- Obtain patches only through pre-approved mirrors, to prevent direct import of urgent patches of unclear origin. +- Include temporary mitigations until the patch is applied (blocking affected functions, restricting access, and so on) in the procedure. +- Set response deadlines according to the severity and exposure of the vulnerability, so that delays are not left unattended. + +## Self risk assessment for network separation exceptions + +When a network separation exception is applied to research and development work, the automatic +protection of the closed network disappears. In that case, the organization must assess the risk of +open source itself, design controls, and document them. + +The self risk assessment document contains the following. + +- The grounds for judging that the work in scope is for research and development and does not process unique identification information or personal credit information. +- The list of open source used in the network separation exception zone (SBOM) and its vulnerability and license risks. +- The record of applying the information protection controls that substitute for network separation, as defined by the Governor of the Financial Supervisory Service. +- The additional risks from opening an internet connection and the security controls that address them (controls that replace import verification). +- How implementation of the controls is confirmed and the reassessment interval. + +Approval of a network separation exception falls under company-wide security governance, not the open +source organization. This document is approved by the Information Protection Committee (or the Chief +Information Security Officer, CISO). The approved assessment becomes supporting material for the open +source use approval stage (section 4) and is subject to periodic reassessment (section 5). +For a form, see the [self risk assessment template for network separation +exceptions](../artifacts/2-policy-templates/#self-risk-assessment-report-for-network-separation-exceptions) provided in the artifacts. + +{{% alert title="Applying to third parties and outsourcing" color="info" %}} +The same import control applies when an electronic financial auxiliary service provider (전자금융보조업자) or an +outsourced development company works inside the closed network or when their artifacts are imported. +How to require an SBOM and vulnerability check results for outsourced artifacts is covered in +identification (section 2) and use approval (section 4). +{{% /alert %}} + +## Connection to the FSEC guide and ISO standards + +Closed network operation is a precondition that comes before the identification and management stages +of the FSEC guide, and at the same time it determines the quality of both stages. The closed network +procedures on this page connect to standard verification materials, FSEC guide procedures, and +regulations as follows. + +| Closed network procedure | Related standard or regulation | FSEC guide procedure | +|------|------|------| +| SBOM generation at the import stage | ISO/IEC 5230 §3.3.1 SBOM management | Identification | +| Fixing components in the internal mirror | ISO/IEC 5230 §3.3.1 Component records | Identification | +| Offline vulnerability management | ISO/IEC 18974 Vulnerability detection and resolution | Issue identification and resolution | +| Managing patch delays | ISO/IEC 18974 Vulnerability remediation | Issue identification and resolution / Management | +| Self risk assessment for a network separation exception | Voluntary security under the Regulation on Supervision of Electronic Financial Transactions (2025-02-05) | Use approval | + +The full mapping of each standard's verification materials to the FSEC guide procedures is in the +[mapping table in the guide overview](../). + +{{% alert title="Field case: KakaoBank's compliance automation" color="success" %}} +At the 12th KWG regular meeting (2021-12), KakaoBank shared a case of automation that moves compliance +checks forward into the development stage. Because checking tools provided only as web services are hard +to use on the internal network under financial sector security requirements, it proposed responding with +command-line tools. This case shows the tool constraints of a closed network. + +Source: Heonkwan Ha (KakaoBank), "Shift-left and Automate Compliance Checks", [KWG 12th meeting (2021-12) slides](https://github.com/OpenChain-Project/OpenChain-KWG/releases/download/meeting-slides-2021/Shift-Left_and_Automate_Compliance_Checks.pdf). +{{% /alert %}} + +--- + +*Last reviewed: 2026-06-10. This page is reviewed when regulations change and once a year on a regular basis.* diff --git a/content/en/guide/finance-oss-guide/1-governance/_index.md b/content/en/guide/finance-oss-guide/1-governance/_index.md new file mode 100644 index 0000000000..994d9af5bd --- /dev/null +++ b/content/en/guide/finance-oss-guide/1-governance/_index.md @@ -0,0 +1,133 @@ +--- +title: "Governance: Management Organization and Approval System" +linkTitle: "1. Governance" +weight: 20 +type: docs +categories: ["guide"] +tags: ["Finance", "Open Source", "Governance", "OSPO", "OSRB"] +description: > + Covers how to form a financial sector open source management organization (OSPO) and review board (OSRB) and divide their roles, + and how to establish approval governance in which legal, security, and technology functions take part together. +--- + +{{% alert title="Where this page fits" color="info" %}} +Governance is the foundation of open source management. The identification, use approval, and +management stages work only after you decide who is responsible, who reviews, and who approves. The +FSEC guide also covers the composition and roles of the management organization separately, before the +management procedures. + +Documents produced here: an open source policy, a list of roles and responsibilities, and records of role assignments. +{{% /alert %}} + +## Where does financial sector governance start? + +The "Guide to Using and Managing Open Source Software in the Financial Sector" (금융분야 오픈소스 +소프트웨어 활용·관리 안내서; Financial Supervisory Service (금융감독원) and Financial Security Institute +(금융보안원), 2022, hereafter the FSEC guide) presents the composition and +operation of the management organization and examples of roles. ISO/IEC 5230 requires these as +verification materials: a documented policy (3.1.1.1), a list of roles and responsibilities (3.1.2.1), +identification of the people in each role (3.2.2.1), and a method for accessing legal expertise +(3.2.2.3). Governance is the activity that produces these verification materials. + +The financial sector adds two more conditions. First, in large organizations where legal, security, +and technology functions are separate, it must be clear who makes open source decisions. Second, in +preparation for supervisory examinations and internal and external audits, the grounds for decisions +and where accountability lies must be kept on record. + +## Open Source Program Office (OSPO) + +The Open Source Program Office (OSPO) is the organization that oversees the use and management of open +source. Some large companies have a dedicated department, and others assign the role to existing +security or development organizations in addition to their current duties. Whatever the size, +responsibility and authority must be gathered in one place. **[FSEC guide]** + +The OSPO's roles are as follows. + +- Establish and update the open source policy. +- Run the open source use approval procedure and convene the review board. +- Take responsibility for the tools and procedures used for identification, vulnerability, and license management. +- Manage the evidence needed for audits and supervisory examinations. +- Coordinate discussions among the legal, security, and development departments. + +Once roles are defined, document the name, contact information, and assignment date of each role +holder. This becomes the verification material for identifying role holders in ISO/IEC 5230 (3.2.2.1) +and for the participant list and role mapping in ISO/IEC 18974 (4.1.2.3). +**[ISO requirement]** + +{{% alert title="What a new organization should do first, and how an operating organization can advance" color="success" %}} +An organization setting up a system for the first time will find it hard to have a dedicated department. +Start with a minimal setup: appoint a single person responsible for open source management, and have +one person each from legal, security, and development take part in addition to their current duties. +Documenting the roles and role holders alone meets the ISO verification material requirements. + +An organization that already operates a system defines the competence required for each role, reviews +the organizational structure and roles regularly while keeping a change history, and measures the +effectiveness of the program with performance indicators. +{{% /alert %}} + +## Open Source Review Board (OSRB) + +The Open Source Review Board (OSRB) is a consultative body that reviews and approves the adoption and +use of open source. It is especially important in the financial sector because a decision to adopt open +source involves license risk, security vulnerabilities, and operational impact together, which makes it +hard for one department to decide alone. + +The board usually looks at the following perspectives together. + +- Legal: license obligations and contract and intellectual property risks. +- Security: vulnerabilities, supply chain risk, and maintenance status. +- Technology and development: technical fit, availability of alternatives, and operational burden. + +Keep records of the board's decisions and their grounds. These records are the core evidence for the +use approval stage (section 4) and are used again in audit response (section 5). If it is hard to +convene the board every time, differentiate the review by risk level. The specific method is covered in +[Use Approval](../4-approve/). **[Guide recommendation]** + +## Legal advice and budget + +ISO/IEC 5230 requires, as verification materials, a method for accessing legal expertise (3.2.2.3) and +confirmation of staffing and budget (3.2.2.2). Financial institutions often have an in-house legal +department, so decide and document the route by which open source license matters are escalated to +legal. Also define the criteria for deciding which matters need external expert advice. +**[ISO requirement]** + +Budget goes to building and operating tools, training, and external advice. A new organization can +lower its initial costs by taking advantage of the fact that most open source tools are themselves open +source and can be installed on premises. Tool selection is covered in [Closed Network +Operation](../0-closed-network/#choosing-tools-for-a-closed-network) and [Management](../5-manage/). + +{{% alert title="Applying to third parties and outsourcing" color="info" %}} +Open source included in deliverables made by electronic financial auxiliary service providers (전자금융보조업자) or +outsourced development companies is also within the scope of governance. +The board decides whether to include open source management requirements (SBOM (Software Bill of +Materials) submission, license notices, vulnerability response) in outsourcing contracts +and who bears that responsibility. Contract and request for proposal management is covered in detail in +[Use Approval](../4-approve/). +{{% /alert %}} + +## Connection to the FSEC guide and ISO standards + +| Governance activity | ISO/IEC 5230 | ISO/IEC 18974 | FSEC guide | +|------|------|------|------| +| Establishing and communicating the policy | 3.1.1.1 Policy, 3.1.1.2 Communication procedure | - | Governance | +| Defining roles and responsibilities | 3.1.2.1 List of roles and responsibilities | 4.1.2.3 Participants and role mapping | Governance | +| Identifying role holders | 3.2.2.1 Document of role holders | - | Governance | +| Access to legal expertise | 3.2.2.3 Method for legal advice | - | Governance | +| Budget and staffing | 3.2.2.2 Confirmation of staffing and budget | - | Other | +| Reviewing and correcting nonconformance | 3.2.2.5 Nonconformance review procedure | - | Other | + +General practice for organization and roles is covered in more detail in the [organization section of the +existing Enterprise Open Source Management Guide](../../opensource_for_enterprise/1-teams/). +This page adds the financial sector's review board and audit-ready records on top of it. + +{{% alert title="Field case: KakaoBank's ISO/IEC 5230 certification" color="success" %}} +At the 13th KWG regular meeting (2022-03), KakaoBank shared its ISO/IEC 5230 certification case. It is a +case of building governance that set up an open source management system in the financial regulatory +environment and went on to achieve certification. + +Source: Heonkwan Ha and Minae Lee (KakaoBank), KakaoBank's part of the 13th meeting joint session "ISO/IEC 5230 Certification Cases of Kakao and KakaoBank" (카카오와 카카오뱅크의 ISO/IEC 5230 인증 사례 공유), [KWG 13th meeting (2022-03) slides](https://github.com/OpenChain-Project/OpenChain-KWG/releases/download/meeting-slides-2022/KakaoBank_ISO_IEC_5230_certification_case.pdf). +{{% /alert %}} + +--- + +*Last reviewed: 2026-06-10. This page is reviewed when regulations change and once a year on a regular basis.* diff --git a/content/en/guide/finance-oss-guide/2-identify/_index.md b/content/en/guide/finance-oss-guide/2-identify/_index.md new file mode 100644 index 0000000000..eb5625071a --- /dev/null +++ b/content/en/guide/finance-oss-guide/2-identify/_index.md @@ -0,0 +1,139 @@ +--- +title: "Identification: Finding All the Open Source You Use" +linkTitle: "2. Identification" +weight: 30 +type: docs +categories: ["guide"] +tags: ["Finance", "Open Source", "SBOM", "Identification", "Supply Chain"] +description: > + Covers how to identify, without omission, newly incoming open source, legacy already in operation, and outsourced deliverables and + record them as SBOMs, and how to integrate with the financial sector supply chain security platform. +--- + +{{% alert title="Where this page fits" color="info" %}} +Identification is the starting point of management. If you do not know what you use, you cannot manage +its vulnerabilities or license obligations. It is the first procedure in the "Guide to Using and Managing Open Source Software in the Financial +Sector" (금융분야 오픈소스 소프트웨어 활용·관리 안내서; Financial Supervisory Service (금융감독원) and +Financial Security Institute (금융보안원), 2022, hereafter the FSEC guide) and the activity +that SBOM management in ISO/IEC 5230 (3.3.1) requires. + +Documents produced here: SBOMs and an inventory of production assets. +{{% /alert %}} + +## What do you identify? + +Identifying open source means finding all three kinds. + +- Newly incoming open source: libraries that developers newly adopt, and their dependencies. +- Open source in systems already in operation: legacy that was not recorded at the time of adoption. +- Open source included in deliverables made by outsourced development companies or electronic financial + auxiliary service providers (전자금융보조업자). An outsourced development company (a contractor entrusted with the + work) and an electronic financial auxiliary service provider (a party that assists a financial institution with + electronic financial transactions under the Electronic Financial Transactions Act (전자금융거래법)) are + legally different categories, but in both cases the open source in their deliverables must be identified. + +You must expand not only the dependencies you declare directly but also the transitive dependencies +they pull in. Most of the actual risk hides in lower-level dependencies that you do not see directly. + +In the financial sector, a regulatory basis applies here as well. Before the 2025-02-05 amendment, +Article 21 of the Regulation on Supervision of Electronic Financial Transactions (전자금융감독규정) +stated that ownership, copyright, and intellectual property rights in products must be made clear when +building information processing systems and in electronic financial transaction contracts, and the FSEC +guide uses this as a basis for its guidance. The amended current Article 21 requires establishing and +operating internal control procedures for contracts, including this. Identifying the origin and license +of open source is the first step toward meeting this requirement. **[FSEC guide]** + +## Creating and recording SBOMs + +Record identification results as an SBOM (Software Bill of Materials). An SBOM is a list of all the open +source components that make up the software, with their versions, licenses, and dependency +relationships. SPDX and CycloneDX are widely used as standard formats. **[ISO requirement]** + +To create SBOMs, you can use open source tools such as [Syft](../../tools/6-syft/), [cdxgen](../../tools/5-cdxgen/), and +[OSV-SCALIBR](../../tools/4-osvscalibr/). All of them can be installed on premises and produce output in +standard formats. The full list of tools is on the [tool pages](../../tools/). The procedure for creating and managing SBOMs becomes the verification material for SBOM management (3.3.1.1) and component records (3.3.1.2) +in ISO/IEC 5230. + +{{% alert title="Applying in a closed network" color="info" %}} +In a closed network, create the SBOM at the stage of importing open source and register it in the +internal mirror together with the open source. This way, what is present and in which version can be seen +in one place. The import procedure and internal mirror setup are covered in +[Closed Network Operation](../0-closed-network/#import-control). +{{% /alert %}} + +## Identifying systems already in operation + +Managing only new adoptions covers only half the picture. In-house systems and servers already in +operation also contain open source adopted long ago, and much of it has no adoption record. Identifying +this legacy with SBOMs is especially important in the financial sector, because vulnerabilities in +production systems are a constant risk. + +Identifying systems in operation is not a one-time task. Build an asset inventory and keep updating +which open source is in which system. This inventory becomes the basis for continuous monitoring +([Management](../5-manage/)). **[Guide recommendation]** + +## Identifying third-party and outsourced deliverables + +The FSEC guide includes the identification of open source used by electronic financial auxiliary +service providers as a check item. The financial sector relies heavily on outsourced development and +entrustment, so open source that you did not adopt directly comes in through outsourced deliverables. If +it is not identified, there is a gap in the scope of management. **[FSEC guide]** + +The method is to require SBOM submission for outsourced deliverables. Make providing an SBOM an obligation +at the contract stage, and verify the SBOMs you receive. The specific requirements to include in +contracts and requests for proposal are covered in [Use Approval](../4-approve/). + +{{% alert title="Applying to third parties and outsourcing" color="info" %}} +Do not trust an SBOM submitted by an outsourcing company as is; verify it. Check again with checking +tools whether the SBOM you received matches the actual deliverable and whether any components are missing. +If the outsourcing company does not have the capability to create SBOMs, also scan the deliverable +yourself to generate an SBOM. +{{% /alert %}} + +## Integration with the supply chain security platform + +The Financial Security Institute's Financial Sector Software Supply Chain Security Platform (금융권 +소프트웨어 공급망 보안 플랫폼) enters full operation in 2026 (for an overview, see the +[guide overview](../)). To participate in the platform, financial institutions should prepare in advance +the format for producing and submitting SBOMs. + +Abroad, the United States and Europe are moving toward mandatory SBOM submission, and the Korean +government has also set out a direction for step-by-step institutionalization through the "Software +Supply Chain Security Guidelines 1.0" (SW 공급망 보안 가이드라인 1.0) (2024-05-13). The government is +pursuing mandatory SBOM submission in the public sector with a target of 2027, and this trend is +expected to spread to the financial sector. SBOM submission is not currently mandatory across the board +for the Korean financial sector, but it is reasonable to prepare for this trend in advance. +**[Guide recommendation]** + +{{% alert title="What a new organization should do first, and how an operating organization can advance" color="success" %}} +An organization setting up a system for the first time starts by creating and recording SBOMs for newly +adopted open source, and chooses one standard format (SPDX or CycloneDX) to use consistently. + +An organization that already operates a system extends the scope of identification to legacy systems +through an asset inventory, requires SBOM submission for outsourced deliverables, and automates SBOM +production to match the submission format of the supply chain security platform. +{{% /alert %}} + +## Connection to the FSEC guide and ISO standards + +| Identification activity | ISO/IEC 5230 | FSEC guide | +|------|------|------| +| Defining scope | 3.1.4.1 Program scope | Identification | +| Creating and managing SBOMs | 3.3.1.1 SBOM management procedure | Identification | +| Recording components | 3.3.1.2 Open source component records | Identification | +| Identifying third-party and outsourced components | 3.3.1.1 applied accordingly | Identification (electronic financial auxiliary service providers) | + +General practice for creating SBOMs and using tools is covered in more detail in the [tools section of the existing Enterprise Open Source Management Guide](../../opensource_for_enterprise/4-tool/) and +the [SBOM clause of the ISO/IEC 5230 Conformance Guide](../../iso5230_guide/3-content-review/1-sbom/). + +{{% alert title="Field case: KFTC's ISO/IEC 5230 conformance" color="success" %}} +At the 25th KWG regular meeting (2025-03), the Korea Financial Telecommunications and Clearings Institute +(금융결제원, KFTC) shared a case of conforming to ISO/IEC 5230:2020 for its payment infrastructure. It is +a case of identifying the open source in financial payment systems and applying a management system. + +Source: Daeyeol Ryu (KFTC), "ISO/IEC 5230:2020 Conformance in KFTC" (금융결제원의 ISO/IEC 5230:2020 준수 사례), [KWG 25th meeting (2025-03) slides](https://github.com/OpenChain-Project/OpenChain-KWG/releases/download/meeting-slides-legacy/20250325_KFTC_OpenChain-KWG_25th_ko.pdf). +{{% /alert %}} + +--- + +*Last reviewed: 2026-06-10. This page is reviewed when regulations change and once a year on a regular basis.* diff --git a/content/en/guide/finance-oss-guide/3-issue-resolve/_index.md b/content/en/guide/finance-oss-guide/3-issue-resolve/_index.md new file mode 100644 index 0000000000..6746a527ab --- /dev/null +++ b/content/en/guide/finance-oss-guide/3-issue-resolve/_index.md @@ -0,0 +1,185 @@ +--- +title: "Identifying and Resolving Issues: Vulnerabilities and Licenses" +linkTitle: "3. Identify and Resolve Issues" +weight: 40 +type: docs +categories: ["guide"] +tags: ["Finance", "Open Source", "Vulnerability", "CVE", "License"] +description: > + Covers the procedure for detecting, assessing, and remediating vulnerabilities in identified open + source and keeping records of it, how to resolve license issues, and how to manage patch delays in + a closed network. +--- + +{{% alert title="Where this page fits" color="info" %}} +Once identification has shown what you use, the next step is to resolve the issues that come out of +it. There are two kinds of issues: security vulnerabilities and license obligations. This is the +second procedure of the FSEC guide, and it corresponds to vulnerability detection and resolution +(4.3.2) in ISO/IEC 18974 and to handling license use cases (3.3.2) in ISO/IEC 5230. + +Documents produced here: vulnerability remediation records (including VEX), and response deadline +criteria by severity. +{{% /alert %}} + +## From vulnerability detection to remediation + +ISO/IEC 18974 requires security assurance activities to be applied to each open source component in +the SBOM (Software Bill of Materials). The whole process from detection to resolution must be set up +as a procedure, and a record of the work done on each vulnerability must be kept. The procedure +follows the flow below. **[ISO requirement]** + +- Detection: Check whether each component in the SBOM has known vulnerabilities (CVE, Common + Vulnerabilities and Exposures). +- Assessment: Assign a risk and impact score to each detected vulnerability, using the Common + Vulnerability Scoring System (CVSS). +- Remediation decision: Decide on and document the fix or mitigation steps needed for each + vulnerability. When you judge that no action is needed, record the basis for that judgment as well. +- Remediation: Apply patches, version replacements, mitigating configurations, and so on according + to the risk score. +- Ongoing response: Monitor vulnerabilities newly disclosed during operation and respond for the + affected systems. + +```mermaid +flowchart LR + A["Detection
(match against SBOM)"] --> B["Assessment
(CVSS score)"] + B --> C{"Is action
needed?"} + C -- Yes --> D["Remediation
patch, replace, mitigate"] + C -- No --> E["Record basis for no action
(VEX)"] + D --> F["Remediation record"] + E --> G["Ongoing response
(monitor new vulnerabilities)"] + F --> G + G --> A + + style A fill:#2b6cb0,color:#fff + style D fill:#c53030,color:#fff + style E fill:#744210,color:#fff + style F fill:#276749,color:#fff + style G fill:#2d3748,color:#fff +``` + +This procedure is the vulnerability detection and resolution procedure of ISO/IEC 18974 (4.3.2.1), +and the record of carrying it out is the vulnerability and remediation record (4.3.2.2). If you +record a decision that no action is needed in VEX (Vulnerability Exploitability eXchange) format, +you do not have to review the same vulnerability again and again. + +### Vulnerability scanning and continuous monitoring example + +When you register an SBOM in a vulnerability management tool, the tool reports the affected +components again each time a new vulnerability is disclosed. The example below registers the SBOM +of an operational system in [Dependency-Track](../../tools/7-dependency-track/). The tool is an +example and can be replaced with an equivalent tool. + +```bash +# Upload the SBOM to a Dependency-Track project (API example) +curl -X POST "https://dtrack.internal/api/v1/bom" \ + -H "X-Api-Key: $DT_APIKEY" \ + -F "project=$PROJECT_UUID" \ + -F "bom=@foo-1.2.3.sbom.json" +``` + +After the upload, the tool matches the SBOM against its vulnerability database and builds a list of +vulnerabilities. In a closed network, this database is updated offline. This is not a one-time +check: each time the database is updated, the SBOMs already registered are reassessed and new +vulnerabilities surface automatically. Continuous monitoring of operational systems is covered +further in [Management](../5-manage/). + +For a one-time check from the command line, you can use [Grype](https://github.com/anchore/grype) +or [Trivy](https://trivy.dev/). + +```bash +# Scan for vulnerabilities using the SBOM as input +grype sbom:foo-1.2.3.sbom.json +``` + +## Response deadlines + +Detection alone is not enough. Unless you set deadlines for remediation by severity, delays are left +unaddressed. Prioritize by looking at severity and exposure together, and state the response +deadline for each level in your policy. The table below is an example of how to set deadlines; +adjust it to your organization's risk tolerance. **[Guide recommendation]** + +| Severity (CVSS) | Internet-facing systems | Internal systems | +|------|------|------| +| Critical (9.0 to 10.0) | Immediately to within a few days | Within a few days to 1 week | +| High (7.0 to 8.9) | Within 1 week | Within 2 weeks to 1 month | +| Medium (4.0 to 6.9) | Within the regular review cycle | Within the regular review cycle | + +Also define temporary mitigations and an exception approval procedure for cases where remediation +within the deadline is difficult. + +{{% alert title="Applying in a closed network: managing patch delays" color="info" %}} +In a closed network, patches also have to go through the import procedure, so they cannot be applied +right away. How to obtain patches only from pre-approved mirrors and how to build temporary +mitigations until a patch is applied into the procedure are covered in [Closed Network +Operation](../0-closed-network/#managing-patch-delays). +{{% /alert %}} + +## Resolving license issues + +The other issue to resolve alongside vulnerabilities is licensing. Check the license obligations of +the identified open source, find uses that do not meet those obligations, and resolve them. +ISO/IEC 5230 requires this as the license use case handling procedure (3.3.2.1). +**[ISO requirement]** + +In the financial sector, the nature of license issues depends on whether the software is +distributed. This guide looks at the two cases separately. + +- Software distributed externally (external services, customer apps): Obligations of + distribution-triggered licenses such as the GPL family (notices, and in some cases source code + disclosure) apply. You need a procedure for meeting license obligations. +- Internal operational systems that are not distributed externally: Because there is no + distribution, source disclosure obligations under the GPL and similar licenses are weak, but you + still check license compatibility and terms of use. + +The FSEC guide also lists, as a check item, having a source code disclosure policy for using +GPL-family software in external distribution. The distinction between distributed software and +internal operational systems is covered further in [Management](../5-manage/). + +For license checks, you can use open source tools such as [FOSSology](../../tools/1-fossology/) and +[SCANOSS](../../tools/9-scanoss/). However, check the license of the tool itself as well. The case of +FOSSLight (AGPL-3.0) is covered in [Tool selection in Closed Network +Operation](../0-closed-network/#choosing-tools-for-a-closed-network). + +{{% alert title="Applying to third parties and outsourcing" color="info" %}} +For vulnerabilities and license issues found in outsourced deliverables, first determine who is +responsible. Check whether the contract specifies the obligation to respond to vulnerabilities and +to fulfill license obligations, and decide how to handle cases where the vendor cannot respond. +Requirements at the contract stage are covered in [Usage Approval](../4-approve/). +{{% /alert %}} + +{{% alert title="What a new organization should do first, and how an operating organization can advance" color="success" %}} +Organizations setting up a program for the first time start by checking and responding to critical +vulnerabilities in internet-facing systems, and by checking the license obligations of the software +they distribute. Start where the risk is highest and narrow down from there. + +Organizations that already run a program put all eight vulnerability handling methods (threat +identification, vulnerability detection, follow-up action, customer notification, analysis of new +vulnerabilities after release, continuous security testing, verification of risk resolution, and +reporting of risk information) in place as procedures, automate vulnerability response, and reuse +decision records through VEX. +{{% /alert %}} + +## Connection to the FSEC guide and ISO standards + +| Issue resolution activity | ISO/IEC 5230 | ISO/IEC 18974 | FSEC guide | +|------|------|------|------| +| Defining vulnerability handling methods | - | 4.1.5.1 Eight handling methods | Identifying and resolving issues | +| Vulnerability detection and resolution procedure | - | 4.3.2.1 Detection and resolution procedure | Identifying and resolving issues | +| Vulnerability and remediation records | - | 4.3.2.2 Vulnerability and remediation records | Identifying and resolving issues | +| Handling license use cases | 3.3.2.1 Handling license use cases | - | Identifying and resolving issues | + +Clause-by-clause details of the vulnerability response procedure are covered in [the security +assurance clauses of the ISO/IEC 18974 Conformance Guide](../../iso18974_guide/3-content-review/2-security-assurance/), +and the eight handling methods in [the standard practice implementation clauses](../../iso18974_guide/1-program-foundation/5-standard-practice/). + +{{% alert title="Field case: KakaoBank's preparation for security assurance" color="success" %}} +At the 20th KWG regular meeting (2023-11), KakaoBank shared how it prepared for ISO/IEC 18974 open +source security assurance. It is a security assurance case from the financial sector in which +vulnerability detection and response were systematized. + +Source: Heonkwan Ha and Minae Lee (KakaoBank), "KakaoBank Open Source Security Assurance Preparation Case Study" (카카오뱅크 오픈소스 보안 보증 준비 사례 공유), [KWG 20th meeting (2023-11) slides](https://github.com/OpenChain-Project/OpenChain-KWG/releases/download/meeting-slides-legacy/kakaobank-ISO18974-conformance-case-study.pdf). +{{% /alert %}} + +--- + +*Last reviewed: 2026-06-10. This page is reviewed when regulations change and once a year on a regular basis.* diff --git a/content/en/guide/finance-oss-guide/4-approve/_index.md b/content/en/guide/finance-oss-guide/4-approve/_index.md new file mode 100644 index 0000000000..5ee74ae067 --- /dev/null +++ b/content/en/guide/finance-oss-guide/4-approve/_index.md @@ -0,0 +1,163 @@ +--- +title: "Usage Approval: What to Allow and on What Basis" +linkTitle: "4. Usage Approval" +weight: 50 +type: docs +categories: ["guide"] +tags: ["Finance", "Open Source", "Usage Approval", "Risk Assessment", "Contract"] +description: > + Covers the workflow for approving open source use, the self risk assessment for network separation + exceptions, and the open source requirements to include in outsourcing contracts and requests for + proposal. +--- + +{{% alert title="Where this page fits" color="info" %}} +This is the step where you decide whether to actually use open source that has gone through +identification and issue resolution. Approval should not rest on one person's judgment; it should +follow a defined procedure and leave records. This is the third procedure of the FSEC guide, and it +connects to license obligation review (3.1.5) and governance in ISO/IEC 5230. + +Documents produced here: approval request, review, and decision records, and open source +requirement clauses for outsourcing contracts. +{{% /alert %}} + +## Usage approval workflow + +Usage approval is the procedure that answers the question "May we use this open source for this +purpose?" The basis for the answer comes from the earlier steps. Identification tells you what it +is, issue identification and resolution tell you its vulnerabilities and license obligations, and +you decide on approval based on those results. + +The basic flow is request, review, decision, and record. + +- Request: Submit the open source you intend to use, its purpose, and whether it will be distributed. +- Review: Look at license obligations, security vulnerabilities, and technical fit together. The + reviewer is the Open Source Review Board (OSRB) defined in governance. +- Decision: Choose one of approval, conditional approval (allowed on the condition of mitigating + measures), or rejection. +- Record: Keep the decision and its basis. This record becomes audit evidence. + +```mermaid +flowchart LR + A["Request
(open source, purpose, distribution)"] --> B{"Is the risk
low?"} + B -- "Yes (meets predefined criteria)" --> C["Automatic approval"] + B -- No --> D["OSRB review
legal, security, technical"] + D --> E["Approval"] + D --> F["Conditional approval
(subject to mitigation)"] + D --> G["Rejection"] + C --> H["Record decision and basis
(audit evidence)"] + E --> H + F --> H + G --> H + + style C fill:#276749,color:#fff + style D fill:#2b6cb0,color:#fff + style G fill:#c53030,color:#fff + style H fill:#2d3748,color:#fff +``` + +Use that involves a network separation exception first requires, separately from this workflow, +approval of the self risk assessment report by the Information Security Committee (or CISO). This is +covered in the section below. + +Not every use needs to be reviewed with the same weight. Differentiate the review: approve low-risk +uses automatically using predefined criteria, and send only uses that go into distributed software +or carry significant license obligations to the board. ISO/IEC 5230 requires the license obligation +review procedure (3.1.5.1) as verification material. **[ISO requirement]** + +{{% alert title="What a new organization should do first, and how an operating organization can advance" color="success" %}} +Organizations setting up a program for the first time start by defining the approval request form +and the review criteria. Decide who submits requests, who reviews them, and who gives final approval, +and document it. + +Organizations that already run a program reduce the review workload by setting automatic approval +criteria by risk level, and link approval records with the identified asset inventory and audit +evidence. +{{% /alert %}} + +## Self risk assessment for network separation exceptions + +Under the amendment to the Regulation on Supervision of Electronic Financial Transactions +(전자금융감독규정), effective 2025-02-05, work for research and development purposes that does not +process unique identification information or personal credit information can be granted a network +separation exception after a self risk assessment and the application of alternative information +security controls in place of network separation. Approval of the network separation exception +itself is not the responsibility of the Open Source Review Board but of company-wide security +governance, that is, the Information Security Committee (or the Chief Information Security Officer, +CISO). + +Open source usage approval interlocks with this procedure. The open source risk information that +goes into the risk assessment report (SBOM, vulnerabilities, licenses) comes from the identification +and issue resolution steps, so the open source management organization provides that information +and supports the review. When reviewing usage approval for open source to be used in a network +separation exception zone, also look at the approved risk assessment report as a supporting +document, and record the decision. What the report contains is covered in [Self risk assessment in +Closed Network +Operation](../0-closed-network/#self-risk-assessment-for-network-separation-exceptions). **[Guide recommendation]** + +{{% alert title="Applying in a closed network" color="info" %}} +For open source used inside a closed network, import approval and usage approval are linked. If you +use the verification results from the import step (integrity, malware scan, SBOM, vulnerability +check) in the usage approval review as well, the procedures do not overlap. For the import +procedure, see [Closed Network Operation](../0-closed-network/#import-control). +{{% /alert %}} + +## Outsourcing contracts and requests for proposal + +The financial sector relies heavily on outsourced development and outsourcing arrangements, so usage +approval extends back to the contract stage. To control the open source included in outsourced +deliverables, you need to put open source management requirements into contracts and requests for +proposal in advance. + +The requirements to include in contracts are as follows. + +- SBOM (Software Bill of Materials) submission: Require the vendor to submit a list of the open source included in the deliverables in a standard format. +- Fulfillment of license obligations: Specify responsibility for license notices and for fulfilling obligations. +- Vulnerability response: Define the obligation and deadline for responding when a vulnerability is found. +- Ownership and attribution of rights: Make clear who owns the deliverables and to whom copyright and intellectual property rights belong. + +The last item relates to Article 21 of the Regulation on Supervision of Electronic Financial +Transactions. Before the 2025-02-05 amendment, Article 21 stated that, in building information +processing systems and in electronic financial transaction contracts, the attribution of ownership, +copyright, and intellectual property rights in products must be made clear, and the FSEC guide also +uses this as the basis for its guidance on managing attribution of rights. The amended current +Article 21, instead of listing detailed items, requires establishing and operating internal control +procedures to secure the safety, reliability, and fairness of contracts, and clarifying attribution +of rights remains a valid practice as a core part of that internal control. Stating the attribution +of rights related to open source clearly in outsourcing contracts meets this requirement. +**[FSEC guide]** + +{{% alert title="Applying to third parties and outsourcing" color="info" %}} +Open source used by electronic financial auxiliary service providers (전자금융보조업자) is also kept +within the same approval system. Review and approve the SBOM and vulnerability check results +submitted by the vendor, and state in the contract who bears that responsibility. Also decide how to +compensate when the vendor lacks management capability (scanning directly, requiring alternative +components). +{{% /alert %}} + +## Connection to the FSEC guide and ISO standards + +| Usage approval activity | ISO/IEC 5230 | FSEC guide | +|------|------|------| +| License obligation review | 3.1.5.1 License obligation review procedure | Usage approval | +| Approval decision and record | 3.1.5.1 Review and record procedure (applying 3.3.1.1 approval procedure) | Usage approval | +| Risk assessment for network separation exceptions | - (self-regulated security under the Regulation on Supervision of Electronic Financial Transactions) | Usage approval | +| Attribution of rights in outsourcing contracts | - (internal control under Article 21 of the Regulation on Supervision of Electronic Financial Transactions) | Usage approval (electronic financial auxiliary service providers) | + +General practice for the approval process is covered in more detail in [the process section of the +Enterprise Open Source Management Guide](../../opensource_for_enterprise/3-process/). For policy and +procedure forms, use the [policy and procedure templates](../../templates/) supplemented for the +financial context; the supplemented outlines are provided as the artifact [Financial Policy and +Procedure Templates](../artifacts/2-policy-templates/). + +{{% alert title="Field case: KakaoBank's certification and approval system" color="success" %}} +At the 13th KWG regular meeting (2022-03), KakaoBank shared its ISO/IEC 5230 certification case and +also introduced its system for reviewing and approving open source use. It is a case of establishing +approval governance in the financial sector. + +Source: Heonkwan Ha and Minae Lee (KakaoBank), KakaoBank's part of the 13th meeting joint session "ISO/IEC 5230 Certification Cases of Kakao and KakaoBank" (카카오와 카카오뱅크의 ISO/IEC 5230 인증 사례 공유), [KWG 13th meeting (2022-03) slides](https://github.com/OpenChain-Project/OpenChain-KWG/releases/download/meeting-slides-2022/KakaoBank_ISO_IEC_5230_certification_case.pdf). +{{% /alert %}} + +--- + +*Last reviewed: 2026-06-10. This page is reviewed when regulations change and once a year on a regular basis.* diff --git a/content/en/guide/finance-oss-guide/5-manage/_index.md b/content/en/guide/finance-oss-guide/5-manage/_index.md new file mode 100644 index 0000000000..14fe7398ce --- /dev/null +++ b/content/en/guide/finance-oss-guide/5-manage/_index.md @@ -0,0 +1,182 @@ +--- +title: "Management: Continuous Checks and Audit Readiness for Internal Operational Systems" +linkTitle: "5. Management" +weight: 60 +type: docs +categories: ["guide"] +tags: ["Finance", "Open Source", "Monitoring", "Audit", "Operational Resilience"] +description: > + Covers how to continuously check and monitor open source in internal operational systems that are + not distributed externally, and how to manage risk in the operation stage through periodic + reassessment and audit evidence management. +--- + +{{% alert title="Where this page fits" color="info" %}} +Checking once at adoption and stopping there is not management. Open source risk continues +throughout operation, and new vulnerabilities are disclosed after adoption. This is the management +(monitoring) step, the last procedure of the FSEC guide, and it corresponds to continuous monitoring +in ISO/IEC 18974. It is the step with the greatest weight in the financial sector. + +Documents produced here: periodic reassessment records, and audit evidence packages. +{{% /alert %}} + +## Distributed software and internal operational systems + +The focus of open source management is often placed on software distributed externally, because +distribution triggers license obligations. In the financial sector, however, open source is used +most in internal operational systems that are not distributed externally: core banking systems, +information systems, internal financial management systems, and internal servers. This guide covers +the two scopes separately. + +- Distributed software (external services, customer apps): License obligations (notices, and in some + cases providing source code) are the focus. This is the domain of ISO/IEC 5230. +- Internal operational systems and servers (not distributed): Because there is no external + distribution, source disclosure obligations are weak, but security vulnerabilities and supply chain + risks are larger and constant. Continuous vulnerability checks, asset inventory, and periodic + reassessment are the focus. This is the domain of ISO/IEC 18974 and operational resilience. + +This page concentrates on the second scope, that is, management of internal operational systems, +because this is an emphasis specific to the financial sector. The European Union's Digital +Operational Resilience Act (DORA) requires ICT operational resilience and management of open source +vulnerabilities and patches for the same reason. **[Guide recommendation]** + +## Operational asset inventory + +Continuous checking starts with knowing what is running. Make a list of the open source contained in +the internal systems and servers in operation. Identify with SBOMs (Software Bill of Materials) not +only new adoptions but also legacy systems that have been in operation for a long time. If you miss +legacy systems with no adoption records, the check scope has gaps. + +The operational asset inventory is the collection of SBOMs created in the identification step +([Identification](../2-identify/)), organized by operational system. If you link which system uses +which components, you can find the affected systems right away when a new vulnerability is disclosed. + +## Continuous vulnerability monitoring + +The core of the operation stage is constant monitoring. Set up a structure that traces back to the +affected operational systems each time a new vulnerability (CVE, Common Vulnerabilities and +Exposures) is disclosed. This activity carries the vulnerability detection and resolution procedure +of ISO/IEC 18974 (4.3.2.1) into the operation stage. **[ISO requirement]** + +The method is to register the SBOMs of operational systems in a vulnerability management tool. When +you register SBOMs in [Dependency-Track](../../tools/7-dependency-track/), all registered SBOMs are +reassessed each time the vulnerability database is updated, and new vulnerabilities surface +automatically. Scan servers, containers, and file systems periodically with +[Trivy](https://trivy.dev/). If you run a CI platform such as GitLab in-house, you can set up the +same continuous checks with its built-in dependency scanning, and commercial SCA tools serve the same +role. + +![Dependency-Track project list with SBOMs registered per system](./dt-projects.png) + +*Figure: Dependency-Track project list with SBOMs registered per operational system. When a new +vulnerability is disclosed, the affected systems show up in this list. The screen reuses a capture +from the [cdxgen and Dependency-Track tutorial](../../tools/8-cdxgen-dt/).* + +### Periodic scanning example + +The example below shows a flow for scanning operational servers regularly and keeping the results. +The tool is an example and can be replaced with an equivalent tool. + +```bash +# Scan the operational server's file system and keep the results by date +trivy fs --format json --output /var/log/oss-scan/$(date +%Y%m%d).json /opt/app + +# In a closed network, update the vulnerability DB offline in advance and turn off automatic updates +TRIVY_SKIP_DB_UPDATE=true trivy fs /opt/app +``` + +In a closed network, the vulnerability database is updated offline. The update procedure is covered +in [Closed Network Operation](../0-closed-network/#offline-vulnerability-management). Scan results +are kept as audit evidence, which is covered next. + +{{% alert title="Applying in a closed network" color="info" %}} +Operational systems in a closed network cannot receive real-time vulnerability information, so the +synchronization cycle of the offline vulnerability database is in effect the cycle at which new +vulnerabilities become known. Set the synchronization cycle and the person responsible, and manage +the risks that can arise during the gap before vulnerabilities become known. For the detailed +procedure, see [Closed Network Operation](../0-closed-network/#offline-vulnerability-management). +{{% /alert %}} + +## Periodic reassessment + +If continuous monitoring is automated surveillance, periodic reassessment is an activity in which +people periodically review the checking system itself. Along with regular reviews such as quarterly +or semi-annual ones, make rechecks at the time of system changes part of the procedure. ISO/IEC 18974 +requires periodic review of the program and evidence of changes (4.1.2.5) as verification material. +**[ISO requirement]** + +Systems with a network separation exception update their self risk assessment regularly. Because +they have been released from network separation through regulatory relief, the responsibility of +self-regulated security continues along with the reassessment cycle. The self risk assessment report +is covered in [Usage Approval](../4-approve/#self-risk-assessment-for-network-separation-exceptions). + +## Audit evidence management + +The financial sector must prepare for internal and external audits and for information technology +(IT) inspections by the Financial Supervisory Service (금융감독원). Check records from the management +step are audit evidence in themselves. If you keep records of what was checked when and how each +vulnerability was handled, you do not need to create separate evidence for audits and inspections. + +It is efficient to reuse the ISO verification material structure as audit evidence. The +vulnerability and remediation records required by ISO/IEC 18974 and the creation and retention of +compliance artifacts required by ISO/IEC 5230 (3.4.1.1, 3.4.1.2) serve directly as audit response +materials. Decide where and for how long to keep evidence, and use a recording method that prevents +tampering. **[Guide recommendation]** + +For a checklist of what to keep and a specification of where to keep it, use the [Audit Evidence +List](../artifacts/3-audit-evidence/) provided as an artifact. + +{{% alert title="What a new organization should do first, and how an operating organization can advance" color="success" %}} +Organizations setting up a program for the first time start continuous monitoring by registering +SBOMs for internet-facing operational systems first, and begin by keeping check records in one +place. + +Organizations that already run a program extend the asset inventory to legacy systems, make the +periodic reassessment cycle a formal procedure, share vulnerability information by linking with the +financial sector's supply chain security platform, and manage audit evidence in a tamper-resistant +form. +{{% /alert %}} + +## Connection to the FSEC guide and ISO standards + +| Management activity | ISO/IEC 5230 | ISO/IEC 18974 | FSEC guide | +|------|------|------|------| +| Continuous vulnerability monitoring | - | 4.3.2 Post-release monitoring | Management | +| Periodic reassessment | - | 4.1.2.5 Periodic review and evidence of changes | Management | +| Creating and retaining artifacts | 3.4.1.1 Creation, 3.4.1.2 Retention | - | Management | +| Audit evidence | Applying 3.4.1.2 Retention | 4.3.2.2 Vulnerability and remediation records | Management | + +Clause-by-clause details of continuous monitoring and security assurance are covered in the +[ISO/IEC 18974 Conformance Guide](../../iso18974_guide/). + +{{% alert title="Field case: KakaoBank's audit response and use of AI" color="success" %}} +At the 30th KWG regular meeting (2026-06), KakaoBank shared its experience responding to audits in +the financial sector. It has responded to inspections and data requests from several institutions, +including the Financial Supervisory Service's IT risk quantitative assessment and management +condition assessment, and the Bank of Korea (한국은행)'s annual survey on the status of financial +informatization. What the institutions commonly checked came down to five things: whether the +open source inventory is managed, the full inventory list, the composition of management staff, +policies and internal rules, and the security vulnerability check procedure. The everyday records of +quarterly self-checks and continuous vulnerability monitoring served directly as response materials. +The tooling is not a single dedicated platform but a combination: KakaoBank said it uses a +commercial tool (FossID) and its own command-line tool for identification and license analysis, +GitLab's dependency scanning for vulnerability checks, and a separate tool (Insight) for SBOM +management. This shows that a combination suited to the organization's environment is realistic. + +At the same meeting, KakaoBank also introduced a case of using an AI coding agent to automate +license identification and compatibility analysis, SBOM generation, and notice writing, which +reduced verification work that had been repeated for every release. + +![Presentation slide on using an AI coding agent for open source governance](./30th-s1-claude-code.png) + +*Figure: Using an AI coding agent for open source governance. Heonkwan Ha (KakaoBank), slides, page 7. +The slide image is the presenter's work and is not covered by the CC BY 4.0 license that applies to +this text.* + +Sources: Minae Lee (KakaoBank), "Experience Report on Open Source Operations as a Financial Company" (금융회사로서의 오픈소스 관련 업무 대응 후기), [slides](https://github.com/OpenChain-Project/OpenChain-KWG/releases/download/meeting-slides-2026/30th-session3-finance-oss-report.pdf); +Heonkwan Ha (KakaoBank), "AI-driven Open Source Governance", [slides](https://github.com/OpenChain-Project/OpenChain-KWG/releases/download/meeting-slides-2026/30th-session1-ai-driven-oss-governance.pdf). KWG 30th meeting (2026-06). +{{% /alert %}} + +--- + +*Last reviewed: 2026-06-10. This page is reviewed when regulations change and once a year on a regular basis.* diff --git a/content/en/guide/finance-oss-guide/6-self-check/_index.md b/content/en/guide/finance-oss-guide/6-self-check/_index.md new file mode 100644 index 0000000000..09c49570dc --- /dev/null +++ b/content/en/guide/finance-oss-guide/6-self-check/_index.md @@ -0,0 +1,123 @@ +--- +title: "Self-Check: Finding the Gaps in Your Program" +linkTitle: "6. Self-Check" +weight: 70 +type: docs +categories: ["guide"] +tags: ["Finance", "Open Source", "Self-Check", "Checklist", "Verification Materials"] +description: > + Links check items, newly written in this guide's own words with reference to the intent of the + FSEC guide's five categories, to ISO verification materials, recommended tools, and guide sections, + so that you can find what is missing from your organization's open source management program. +--- + +{{% alert title="Where this page fits" color="info" %}} +This page gathers the preceding six steps into a checklist. Go through the items one by one to find +the gaps in your program, and move to the section that covers what is missing. The categories are +based on the self-check checklist (Appendix 1) of the FSEC guide. + +Documents produced here: the self-check workbook (check results and improvement plan). +{{% /alert %}} + +## How to use the self-check + +This checklist is not a pass-or-fail test. It is a tool for gauging how far your organization has +come and finding what to do next. Just as the FSEC guide is a non-regulatory voluntary check, this +checklist is a best-practice baseline. **[FSEC guide]** + +Each item is linked to the related ISO verification materials, recommended tools, and the guide +section that covers it in detail. If there is an item you do not meet, go to the linked section and +check how to address it. The recommended tools are examples, and each tool name links to its usage +page at its first appearance. The items follow the order of the five categories used by the FSEC +guide: identification, identifying and resolving issues, approval, management, and other. + +{{% alert title="What a new organization should do first, and how an operating organization can advance" color="success" %}} +Organizations setting up a program for the first time should not try to check every item at once. +Start with the other category, which forms the foundation, and the identification category, which is +the starting point, and decide what to put in place first. + +Organizations that already run a program check every item and then move the unmet items into the +improvement plan (owner, target date) of the [self-check workbook](../artifacts/1-workbook/) to +manage them. +{{% /alert %}} + +{{% alert title="Copyright notice" color="warning" %}} +The check items below are not copied from the checklist questions of the FSEC guide. They were newly +written in this guide's own words with reference to the intent of the five categories. To check the +original guide as well, see [the material published by the Financial Security Institute +(금융보안원)](https://www.fsec.or.kr/bbs/detail?menuNo=222&bbsNo=11166). +{{% /alert %}} + +## Identification + +| Check item | Related ISO verification material | Recommended tools | Guide section | +|------|------|------|------| +| The scope of open source management is defined in a document | 5230 3.1.4.1 | - | [Identification](../2-identify/) | +| Newly adopted open source and its dependencies are fully identified | 5230 3.3.1.1 | [Syft](../../tools/6-syft/), [cdxgen](../../tools/5-cdxgen/) | [Identification](../2-identify/) | +| Transitive dependencies are expanded and identified | 5230 3.3.1.1 | Syft, [OSV-SCALIBR](../../tools/4-osvscalibr/) | [Identification](../2-identify/) | +| Legacy open source in systems already in operation is identified | 5230 3.3.1.1 | [Trivy](https://trivy.dev/), [Dependency-Track](../../tools/7-dependency-track/) | [Identification](../2-identify/#identifying-systems-already-in-operation) | +| Identification results are recorded as SBOMs in a standard format | 5230 3.3.1.2 | cdxgen, Syft | [Identification](../2-identify/#creating-and-recording-sboms) | +| Open source in deliverables from outsourcing vendors and electronic financial auxiliary service providers is identified | Applying 5230 3.3.1.1 | (Require SBOM submission) | [Identification](../2-identify/#identifying-third-party-and-outsourced-deliverables) | + +## Identifying and resolving issues + +| Check item | Related ISO verification material | Recommended tools | Guide section | +|------|------|------|------| +| Each component in the SBOM is checked for known vulnerabilities | 18974 4.3.2.1 | Dependency-Track, [Grype](https://github.com/anchore/grype), Trivy | [Identify and Resolve Issues](../3-issue-resolve/) | +| Vulnerabilities are given risk scores and response deadlines are set by severity | 18974 4.3.2.1 | Dependency-Track, Grype | [Identify and Resolve Issues](../3-issue-resolve/#response-deadlines) | +| Vulnerability remediation results are recorded (including decisions that no action is needed) | 18974 4.3.2.2 | Dependency-Track (VEX), commercial SCA | [Identify and Resolve Issues](../3-issue-resolve/) | +| License obligations are checked and conflicts are resolved | 5230 3.3.2.1 | [FOSSology](../../tools/1-fossology/), [SCANOSS](../../tools/9-scanoss/) | [Identify and Resolve Issues](../3-issue-resolve/#resolving-license-issues) | +| A source code disclosure policy for GPL-family software in external distribution is in place | 5230 3.3.2.1 | - | [Identify and Resolve Issues](../3-issue-resolve/#resolving-license-issues) | + +## Approval + +| Check item | Related ISO verification material | Recommended tools | Guide section | +|------|------|------|------| +| An open source usage approval procedure and reviewer are defined | 5230 3.1.5.1 | [SW360](../../tools/2-sw360/), Dependency-Track (policies) | [Usage Approval](../4-approve/) | +| Approval decisions and their basis are recorded | 5230 3.1.5.1 | SW360, Dependency-Track | [Usage Approval](../4-approve/#usage-approval-workflow) | +| For network separation exceptions, the basis for approval is kept in a self risk assessment report | (Regulation on Supervision of Electronic Financial Transactions (전자금융감독규정)) | - | [Usage Approval](../4-approve/#self-risk-assessment-for-network-separation-exceptions) | +| Outsourcing contracts include SBOM, license, vulnerability, and attribution-of-rights requirements | (Attribution of rights: Article 21 of the Regulation on Supervision of Electronic Financial Transactions; others: Guide recommendation) | - | [Usage Approval](../4-approve/#outsourcing-contracts-and-requests-for-proposal) | + +## Management + +| Check item | Related ISO verification material | Recommended tools | Guide section | +|------|------|------|------| +| SBOMs of operational systems are registered and continuously monitored for vulnerabilities | 18974 4.3.2 | Dependency-Track, CI dependency scanning, commercial SCA | [Management](../5-manage/#continuous-vulnerability-monitoring) | +| When a new vulnerability is disclosed, the affected systems are traced back | 18974 4.3.2 | Dependency-Track, commercial SCA | [Management](../5-manage/#continuous-vulnerability-monitoring) | +| A periodic reassessment cycle is defined | 18974 4.1.2.5 | - | [Management](../5-manage/#periodic-reassessment) | +| Check records are kept as audit evidence | 5230 3.4.1.2 | - | [Management](../5-manage/#audit-evidence-management) | + +## Other + +The other category is the foundation that supports the preceding four steps. The FSEC guide also +places selection criteria, exception approval, role documents, staffing, budget, policy awareness, +and legal advice in a separate category. + +| Check item | Related ISO verification material | Recommended tools | Guide section | +|------|------|------|------| +| Open source selection criteria and an exception approval procedure exist | 5230 3.1.1.1 | - | [Governance](../1-governance/) | +| Roles and responsibilities are documented | 5230 3.1.2.1, 18974 4.1.2.3 | - | [Governance](../1-governance/#open-source-program-office-ospo) | +| Staff and budget are secured | 5230 3.2.2.2 | - | [Governance](../1-governance/#legal-advice-and-budget) | +| The policy is communicated to members of the organization | 5230 3.1.1.2 | - | [Governance](../1-governance/) | +| There is a channel for accessing legal advice | 5230 3.2.2.3 | - | [Governance](../1-governance/#legal-advice-and-budget) | + +## Self-check workbook + +The [self-check workbook](../artifacts/1-workbook/), which combines the check items above with check +result records, ISO verification materials, owners, and target dates in a single sheet, is provided +as an artifact. Check the recommended tools for each item in the tables on this page. + +## Cross-reference to ISO verification materials + +The items in this checklist link to the verification materials of ISO/IEC 5230 and 18974. If you +meet a check item and keep the supporting document, that document serves directly as verification +material for ISO self-certification. If, after checking every item, you write a document confirming +that all requirements are met, that document becomes the verification material for confirming that +the requirements of ISO/IEC 5230 are met (3.6.1.1). Clause-by-clause details of the verification +materials are in the [ISO/IEC 5230 Conformance Guide](../../iso5230_guide/) and the [ISO/IEC 18974 +Conformance Guide](../../iso18974_guide/), and the full mapping is in the [mapping table in the +guide overview](../#mapping-of-the-fsec-guide-iso-standards-and-kwg-guides). + +--- + +*Last reviewed: 2026-06-10. This page is reviewed when regulations change and once a year on a regular basis.* diff --git a/content/en/guide/finance-oss-guide/_index.md b/content/en/guide/finance-oss-guide/_index.md new file mode 100644 index 0000000000..94f484906a --- /dev/null +++ b/content/en/guide/finance-oss-guide/_index.md @@ -0,0 +1,200 @@ +--- +title: "Practical Guide to Open Source Management in the Financial Sector" +linkTitle: "Financial Open Source Guide" +weight: 46 +type: docs +categories: ["guide"] +tags: ["Finance", "Open Source", "Compliance", "Supply Chain Security", "Network Separation"] +description: > + A practical guide for the financial sector that connects the procedures of the Guide to Using and + Managing Open Source Software in the Financial Sector with the international standards (ISO/IEC 5230 and 18974) so that they can be put into actual operation. +--- + +{{% pageinfo %}} + +Financial institutions use more and more open source, yet they must manage it under conditions specific +to their sector: closed networks, network separation regulations, supply chain security, and audit +response. This guide connects the management procedures set out in the "Guide to Using and Managing +Open Source Software in the Financial Sector" (금융분야 오픈소스 소프트웨어 활용·관리 안내서; +Financial Supervisory Service (금융감독원) and Financial Security Institute (금융보안원), 2022) with the +verification materials of the international standards ISO/IEC 5230 and 18974 and with the existing KWG +practical guides. It shows practitioners, in an actionable form, what they need to put in place and in +what order. + +**Author : OpenChain Korea Work Group / [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)** + +{{% /pageinfo %}} + +{{% alert title="Nature and scope of this guide" color="warning" %}} +This guide does not replace the Guide to Using and Managing Open Source Software in the Financial +Sector (hereafter the FSEC guide). It is supplementary material that helps put the procedures defined +in the FSEC guide into actual operation. Just as the FSEC guide is non-regulatory, voluntary guidance, +the recommendations in this guide are good practice, not legal obligations. + +The scope is the same as that of the FSEC guide. It focuses on the use and management of open source +(license compliance and security management) and does not cover publishing in-house software or +contributing to open source projects. This decision reflects the financial sector's reluctance to +publish software. +{{% /alert %}} + +## The regulatory environment of the financial sector + +Open source management in the financial sector takes place in a regulatory environment that differs +from that of ordinary companies. + +The reference for open source management is the FSEC guide. The Financial Supervisory Service and the +Financial Security Institute jointly published it in 2022 (a revised edition from February 2023 also +exists). It sets out minimum security management procedures in four stages: identification, issue +identification and resolution, use approval, and management. It is non-regulatory, voluntary guidance, +and its annexes include a self-check checklist (five categories: identification, issue identification +and resolution, approval, management, and other), major management tools, and operational cases. + +The regulatory environment is in transition. Following the "Roadmap for Improving Network Separation in +the Financial Sector" (금융분야 망분리 개선 로드맵) by the Financial Services Commission (금융위원회) +(2024-08-13), amendments to the Regulation on Supervision of Electronic Financial Transactions +(전자금융감독규정) and its Enforcement Rules (시행세칙) took effect on 2025-02-05. With this, regulation +centered on physical network separation began to shift toward voluntary security and a risk-based +approach, and an exception to network separation was opened for research and development work. The +requirements for applying the exception, further relaxations, and how to respond are covered in +[Closed Network Operation and the Network Separation +Transition](./0-closed-network/). + +Supply chain security is becoming more important. The Financial Security Institute has built the +Financial Sector Software Supply Chain Security Platform (금융권 소프트웨어 공급망 보안 플랫폼), which +went through a pilot operation at the end of 2025 and enters full operation in 2026. It provides +integrated vulnerability management for the financial sector, an SBOM (Software Bill of Materials) +management system, and more efficient bug bounty operation. The government also announced the +"Software Supply Chain Security Guidelines 1.0" (SW 공급망 보안 가이드라인 1.0; Ministry of Science and +ICT (과학기술정보통신부), National Intelligence Service (국가정보원), and Presidential Committee on the +Digital Platform Government (디지털플랫폼정부위원회), 2024-05-13), and is pursuing step-by-step +institutionalization with the goal of mandatory SBOM submission in the public sector in 2027. Abroad, +the United States and Europe are moving toward mandatory SBOM submission, and the European Union's +Digital Operational Resilience Act (DORA) entered into force in 2023 and applies from 2025-01-17. It +requires ICT third-party risk management and management of open source vulnerabilities and patches. It +directly affects financial institutions that operate in or are entering Europe. + +## How to read this guide + +How you read this guide depends on the maturity of your open source management system. Each section +distinguishes what a new organization should do first from how an operating organization can advance, +so assess where your organization stands and then choose the relevant parts. + +An organization setting up a system for the first time sorts out its environment with [Closed Network +Operation](./0-closed-network/), then follows the sections in order from +[Governance](./1-governance/) to [Management](./5-manage/), creating the documents and procedures each +stage describes one at a time. + +An organization that already operates a system evaluates its current state with +[Self-Check](./6-self-check/) and goes back to the stages where it falls short to improve them. + +### Maturity ladder + +The Open Source Readiness maturity model of FINOS (Fintech Open Source Foundation) divides an +organization's stages into +Usage, Compliance, Contribution, Hosting, and Strategic Leverage. Because this guide focuses on use and +management, it corresponds to the first two stages: Usage, where the organization uses open source, and +Compliance, where it has compliance in place. Stages from Contribution onward are not covered. + +| Stage | State | What to do in this guide | FINOS equivalent | +|------|------|------|------| +| New | Uses open source but has no management system | Sort out the closed network environment, establish governance, start identification | Usage | +| Operating | Identification, approval, and management procedures are running | Formalize vulnerability response procedures, systematize use approval, continuous monitoring | Compliance | +| Advanced | Procedures are automated and audits can be answered | Integrate with the supply chain platform, audit evidence system, periodic reassessment | Compliance (advanced) | + +## Mapping of the FSEC guide, ISO standards, and KWG guides + +This table links each procedure in the FSEC guide to the corresponding ISO/IEC 5230 and 18974 +verification materials, to the pages of existing KWG guides where specific methods can be found, and to +the section of this guide that covers it. It is the starting material for self-certification +preparation. + +| FSEC procedure | Section of this guide | ISO/IEC 5230 verification material | ISO/IEC 18974 verification material | Existing KWG guide | +|------|------|------|------|------| +| (Governance) | [1. Governance](./1-governance/) | 3.1.1.1 Policy, 3.1.2.1 Roles and responsibilities, 3.2.2.1 Role holders | 4.1.2.3 Participants and role mapping | [Organization](../opensource_for_enterprise/1-teams/) | +| Identification | [2. Identification](./2-identify/) | 3.1.4.1 Scope, 3.3.1.1 SBOM management, 3.3.1.2 Component records | - | [SBOM and tools](../opensource_for_enterprise/4-tool/), [§3.3.1](../iso5230_guide/3-content-review/1-sbom/) | +| Issue identification and resolution | [3. Issue Identification and Resolution](./3-issue-resolve/) | 3.3.2.1 Handling license use cases | 4.3.2.1 Vulnerability detection and resolution procedure, 4.3.2.2 Records of vulnerabilities and actions | [ISO/IEC 18974 guide](../iso18974_guide/) | +| Use approval | [4. Use Approval](./4-approve/) | 3.1.5.1 License obligation review procedure | - | [Process](../opensource_for_enterprise/3-process/) | +| Management | [5. Management](./5-manage/) | 3.4.1.1 Creating compliance artifacts, 3.4.1.2 Archiving compliance artifacts | 4.1.2.5 Evidence of periodic review and changes | [ISO/IEC 18974 guide](../iso18974_guide/) | +| (Check across all stages) | [6. Self-Check](./6-self-check/) | 3.6.1.1 Confirmation of meeting requirements | - | [Declaring conformance](../opensource_for_enterprise/6-conforming/) | + +ISO/IEC 5230 consists of 25 verification materials across 13 clauses, and ISO/IEC 18974 consists of 25 +verification materials from a security assurance perspective (9 of which are specific to 18974). For +clause-by-clause details of the verification materials, see the [ISO/IEC 5230 Conformance +Guide](../iso5230_guide/) and the [ISO/IEC 18974 Conformance Guide](../iso18974_guide/). + +## Structure of this guide + +Sections 2 to 5 correspond to the four stages of the FSEC guide (identification, issue identification +and resolution, use approval, and management). Before and after them are closed network operation (0) +and governance (1), which are common premises in the financial sector, and self-check (6), which reviews +all stages. + +```mermaid +flowchart LR + P0["0. Closed network operation
(common premise)"] --> P1["1. Governance
(management organization)"] + P1 --> P2["2. Identification"] + P2 --> P3["3. Issue identification
and resolution"] + P3 --> P4["4. Use approval"] + P4 --> P5["5. Management
(continuous monitoring)"] + P5 --> P6["6. Self-check"] + P6 -. "Return to the stage that falls short" .-> P2 + + style P0 fill:#2d3748,color:#fff + style P1 fill:#2d3748,color:#fff + style P2 fill:#2b6cb0,color:#fff + style P3 fill:#2b6cb0,color:#fff + style P4 fill:#2b6cb0,color:#fff + style P5 fill:#2b6cb0,color:#fff + style P6 fill:#276749,color:#fff +``` + +| Section | What it covers | +|------|------| +| [0. Closed Network Operation](./0-closed-network/) | Import control, internal mirrors, offline vulnerability management, self risk assessment for network separation exceptions | +| [1. Governance](./1-governance/) | Management organization and roles, Open Source Review Board (OSRB), approval governance | +| [2. Identification](./2-identify/) | Identifying incoming open source and legacy, SBOM, identifying third-party and outsourced components, integration with the supply chain platform | +| [3. Issue Identification and Resolution](./3-issue-resolve/) | Vulnerability assessment and response, resolving license issues, managing patch delays | +| [4. Use Approval](./4-approve/) | Approval workflow, risk assessment for network separation exceptions, contract (request for proposal) management | +| [5. Management](./5-manage/) | Continuous monitoring of in-house production systems, periodic reassessment, audit evidence | +| [6. Self-Check](./6-self-check/) | Self-check workbook (check items, verification materials, and tool links) | + +## Starting with the self-check + +If you are not sure what stage your organization is at, start with [Self-Check](./6-self-check/). Its +check items were newly written in the wording of this guide, drawing on the intent of the five +categories of the FSEC guide, and are linked to ISO verification materials and recommended tools. You +can use them to find what is missing and move to the section that covers it. + +{{% alert title="Labeling convention" color="info" %}} +Each page distinguishes sources with the following labels. + +- **[ISO requirement]**: An item that the ISO/IEC 5230 or 18974 standard explicitly requires as a verification material. +- **[FSEC guide]**: An item that the FSEC guide sets out as a procedure. The FSEC guide is non-regulatory, voluntary guidance. +- **[Guide recommendation]**: An item not found in the standards or the FSEC guide but recommended on the basis of practice and the financial regulatory environment. + Adoption is at the organization's discretion. +{{% /alert %}} + +{{% alert title="Relationship with other KWG guides" color="success" %}} +This guide builds on the existing guides. General practice for organization, policy, process, tools, and +training is covered in the [Enterprise Open Source Management Guide](../opensource_for_enterprise/), and +clause-by-clause details of the standards' verification materials are covered in the [ISO/IEC 5230 +Conformance Guide](../iso5230_guide/) and the [ISO/IEC 18974 Conformance Guide](../iso18974_guide/). +This guide adds the financial sector context of closed networks, network separation, supply chain, and +audits on top of them. +{{% /alert %}} + +## Sources + +- Guide to Using and Managing Open Source Software in the Financial Sector (Financial Supervisory Service and Financial Security Institute, 2022). [Posted by the Financial Security Institute](https://www.fsec.or.kr/bbs/detail?menuNo=222&bbsNo=11166) +- Roadmap for Improving Network Separation in the Financial Sector (Financial Services Commission, 2024-08-13). [Financial Services Commission press release](https://www.fsc.go.kr/no010101/82885) +- Amendments to the Regulation on Supervision of Electronic Financial Transactions and its Enforcement Rules (Financial Services Commission, effective 2025-02-05). +- Financial Sector Software Supply Chain Security Platform (Financial Security Institute, full operation in 2026). [DailySecu report](https://www.dailysecu.com/news/articleView.html?idxno=204828) +- Software Supply Chain Security Guidelines 1.0 (Ministry of Science and ICT, National Intelligence Service, and Presidential Committee on the Digital Platform Government, 2024-05-13). [Korea Policy Briefing](https://www.korea.kr/news/policyNewsView.do?newsId=148929111) +- Plan for mandatory SBOM submission in the public sector (target and plan for 2027). [Datanet report](https://www.datanet.co.kr/news/articleView.html?idxno=208920) +- EU DORA (Digital Operational Resilience Act, applies from 2025-01-17). [EIOPA overview](https://www.eiopa.europa.eu/digital-operational-resilience-act-dora_en) +- [FINOS Open Source Readiness](https://osr.finos.org/docs/bok/introduction) +- [ISO/IEC 5230](https://www.iso.org/standard/81039.html), [ISO/IEC 18974](https://www.iso.org/standard/86450.html) + +--- + +*Last reviewed: 2026-06-10. This guide is reviewed when regulations change and once a year on a regular basis.* diff --git a/content/en/guide/finance-oss-guide/artifacts/1-workbook/_index.md b/content/en/guide/finance-oss-guide/artifacts/1-workbook/_index.md new file mode 100644 index 0000000000..c475e4022a --- /dev/null +++ b/content/en/guide/finance-oss-guide/artifacts/1-workbook/_index.md @@ -0,0 +1,87 @@ +--- +title: "Self-Check Workbook" +linkTitle: "Self-Check Workbook" +weight: 10 +type: docs +categories: ["guide"] +tags: ["Finance", "Open Source", "Self-Check", "Workbook"] +description: > + A workbook form for recording the check items of the five categories together with compliance + status, supporting documents, owners, and target dates. Use it to turn the items on the self-check + page into an improvement plan. +--- + +This workbook turns the check items on the [Self-Check](../../6-self-check/) page into a form for +recording. Its purpose is to go beyond checking: for the areas that fall short, record the owner and +deadline so that the results lead into an improvement plan. + +## How to use + +Mark the compliance status of each item at one of three levels. + +- Not met: There is no related activity or document. +- Partially met: The activity exists, but documents or records are insufficient. +- Met: Both the activity and the supporting documents are in place. + +For partially met or unmet items, write what the supporting document should be, and who will put it +in place by when. For met items, write the name and location of the supporting document so that it +can be used directly as verification material for ISO self-certification. + +Copy the tables below and fill them in with your organization's check records. In the status column, +enter one of Not met, Partial, or Met. + +## Identification + +| Check item | ISO verification material | Status | Supporting document and location | Owner | Target date | +|------|------|------|------|------|------| +| The scope of open source management is defined in a document | 5230 3.1.4.1 | | | | | +| Newly adopted open source and its dependencies are fully identified | 5230 3.3.1.1 | | | | | +| Transitive dependencies are expanded and identified | 5230 3.3.1.1 | | | | | +| Legacy open source in systems already in operation is identified | 5230 3.3.1.1 | | | | | +| Identification results are recorded as SBOMs in a standard format | 5230 3.3.1.2 | | | | | +| Open source in deliverables from outsourcing vendors and electronic financial auxiliary service providers (전자금융보조업자) is identified | Applying 5230 3.3.1.1 | | | | | + +## Identifying and resolving issues + +| Check item | ISO verification material | Status | Supporting document and location | Owner | Target date | +|------|------|------|------|------|------| +| Known vulnerabilities in each component are checked | 18974 4.3.2.1 | | | | | +| Risk scores are assigned and response deadlines are set | 18974 4.3.2.1 | | | | | +| Vulnerability remediation results are recorded | 18974 4.3.2.2 | | | | | +| License obligations are checked and conflicts are resolved | 5230 3.3.2.1 | | | | | +| A source code disclosure policy for GPL-family software in distribution is in place | 5230 3.3.2.1 | | | | | + +## Approval + +| Check item | ISO verification material | Status | Supporting document and location | Owner | Target date | +|------|------|------|------|------|------| +| A usage approval procedure and reviewer are defined | 5230 3.1.5.1 | | | | | +| Approval decisions and their basis are recorded | 5230 3.1.5.1 | | | | | +| For network separation exceptions, a self risk assessment report is kept | (Regulation on Supervision of Electronic Financial Transactions (전자금융감독규정)) | | | | | +| Outsourcing contracts include open source requirements | (Attribution of rights: Article 21 of the Regulation on Supervision of Electronic Financial Transactions; others: Guide recommendation) | | | | | + +## Management + +| Check item | ISO verification material | Status | Supporting document and location | Owner | Target date | +|------|------|------|------|------|------| +| SBOMs of operational systems are registered and continuously monitored | 18974 4.3.2 | | | | | +| When a new vulnerability is disclosed, the affected systems are traced back | 18974 4.3.2 | | | | | +| A periodic reassessment cycle is defined | 18974 4.1.2.5 | | | | | +| Check records are kept as audit evidence | 5230 3.4.1.2 | | | | | + +## Other + +| Check item | ISO verification material | Status | Supporting document and location | Owner | Target date | +|------|------|------|------|------|------| +| Selection criteria and an exception approval procedure exist | 5230 3.1.1.1 | | | | | +| Roles and responsibilities are documented | 5230 3.1.2.1, 18974 4.1.2.3 | | | | | +| Staff and budget are secured | 5230 3.2.2.2 | | | | | +| The policy is communicated to members of the organization | 5230 3.1.1.2 | | | | | +| There is a channel for accessing legal advice | 5230 3.2.2.3 | | | | | + +The guide sections that cover each check item in detail are linked from the tables on the +[Self-Check](../../6-self-check/) page. + +--- + +*Last reviewed: 2026-06-10. This page is reviewed when regulations change and once a year on a regular basis.* diff --git a/content/en/guide/finance-oss-guide/artifacts/2-policy-templates/_index.md b/content/en/guide/finance-oss-guide/artifacts/2-policy-templates/_index.md new file mode 100644 index 0000000000..8f5174654a --- /dev/null +++ b/content/en/guide/finance-oss-guide/artifacts/2-policy-templates/_index.md @@ -0,0 +1,194 @@ +--- +title: "Policy and Procedure Templates" +linkTitle: "Policy and Procedure Templates" +weight: 20 +type: docs +categories: ["guide"] +tags: ["Finance", "Open Source", "Policy", "Template", "Network Separation"] +description: > + Provides outlines for a finance-specific open source policy, an import procedure, a usage approval + form, and a self risk assessment report for network separation exceptions. Fill them in to suit + your organization. +--- + +The forms here are outlines that add items specific to the financial sector. For general open source +policy and process forms, use the existing [policy and procedure templates](../../../templates/) +guide, and use the forms below for what the financial sector additionally needs: import control, +risk assessment for network separation exceptions, and SBOM requirements for outsourcing. + +{{% alert title="How to use" color="info" %}} +The outlines below are not finished documents to use as they are, but frames to fill in. Replace +the parts in square brackets to suit your organization. Have them reviewed by your internal legal +and security teams before issuing them. +{{% /alert %}} + +## Financial open source policy (supplementary items) + +Add items specific to the financial sector to your existing open source policy. Below is an outline +of the items to include in the policy. + +```text +1. Purpose and scope + - Covers both distributed software and internal operational systems. + - Applies separately to closed networks and network separation exception zones. + +2. Identification + - Identify open source in new, legacy, and outsourced deliverables with SBOMs. + +3. Identifying and resolving issues + - Response deadlines by vulnerability severity: [deadlines set by the organization]. + - Source code disclosure policy for GPL-family software in external distribution: [policy details]. + +4. Usage approval + - Composition and authority of the approver (Open Source Review Board). + - Approval levels by risk level (automatic approval criteria and criteria for referral to the board). + +5. Management + - Continuous monitoring of operational systems and periodic reassessment cycle: [quarterly, semi-annual, etc.]. + - Audit evidence retention period: [period set by the organization]. + +6. Network separation exceptions + - Obligation of self risk assessment and reassessment cycle. + +7. Roles and responsibilities, budget, access to legal advice, policy communication +``` + +## Open source import procedure (closed network) + +This is an outline of the procedure for bringing open source into a closed network. For a detailed +explanation, see [Closed Network Operation](../../0-closed-network/#import-control). + +```text +Import request + - Requester, target open source and version, purpose, reason for import + +Verification in the external zone + - Confirm the official distribution source, compare checksums (integrity) + - Malware scan + - SBOM generation + - Preliminary vulnerability check + +Import approval + - Review verification results, decide on approval + - Record approver and approval date and time + +Internal transfer + - Cross-network data transfer, recheck hashes inside the internal network + - Register in the internal mirror (register the SBOM as well) + +Record retention + - Keep the entire process above as audit evidence +``` + +## Usage approval request and review form + +```text +[Request] + - Requester / Department / Request date + - Target open source / Version / License + - Purpose of use / Distribution (external distribution / internal operation) + - Attachments: SBOM, vulnerability check results + +[Review] (Open Source Review Board) + - Legal: review opinion on license obligations, contracts, and attribution of rights + - Security: review opinion on vulnerabilities, supply chain, and maintenance status + - Technical: review opinion on suitability and replaceability + +[Decision] + - Approval / Conditional approval (conditions: ___) / Rejection + - Decision maker / Decision date / Basis +``` + +{{% alert title="Filled-in example (fictional case)" color="info" %}} +Below is a fictional case showing how to fill in the form. OO Bank and the systems and dates that +appear are all made up and have no relation to any real institution or system. +{{% /alert %}} + +```text +[Request] + - Requester: Kim OO / Digital Channel Development Team / 2026-05-12 + - Target open source: Apache Kafka / 3.9.x / Apache-2.0 + - Purpose of use: internal event streaming (loading transaction notifications) / internal operation (not distributed) + - Attachments: kafka-3.9.sbom.json, vulnerability check results (critical 0, high 0) + +[Review] (Open Source Review Board, 2026-05-19) + - Legal: Apache-2.0, no notice obligation because of non-distributed use. No issues. + - Security: No known critical vulnerabilities. Import via the internal mirror confirmed. No issues. + - Technical: Consistent with the existing messaging standard. Operations team's operating capability confirmed. + +[Decision] + - Conditional approval (condition: register the operational system's SBOM in Dependency-Track and include it in continuous monitoring) + - Decision maker: Open Source Review Board / 2026-05-19 / Synthesis of three review opinions +``` + +## Self risk assessment report for network separation exceptions + +Prepare this report when applying a network separation exception, under the amendment to the +Regulation on Supervision of Electronic Financial Transactions (전자금융감독규정) effective +2025-02-05, to work for research and development purposes that does not process unique +identification information or personal credit information. For a detailed explanation, see [Self +risk assessment in Closed Network +Operation](../../0-closed-network/#self-risk-assessment-for-network-separation-exceptions). + +```text +1. Target work + - Work name / System name + - Whether it qualifies as research and development purposes, and the basis for the judgment + - Confirmation that no unique identification information or personal credit information is processed + +2. Open source in use + - SBOM (list of open source used in the target zone) + - Vulnerability and license risk assessment + +3. Additional risks and controls + - Risks added by the internet connection + - Alternative information security controls applied in place of network separation + - Security controls replacing import verification + +4. Implementation check and reassessment + - How to confirm implementation of controls + - Reassessment cycle: [cycle set by the organization, usually at least once a year] + +5. Review and approval + - Author / Date written + - Review: information security department, open source management organization (confirmation of risk information) + - Approval: Information Security Committee or Chief Information Security Officer (CISO) / Approval date +``` + +{{% alert title="Filled-in example (fictional case)" color="info" %}} +Below is a fictional case showing how to fill in the form. OO Bank and the work and dates that +appear are all made up and have no relation to any real institution or system. +{{% /alert %}} + +```text +1. Target work + - Work name: fraud detection model research / System name: FDS research and validation environment + - Qualifies as research and development purposes: Yes. A dedicated environment for model research and validation, separated from operational services + - No unique identification information or personal credit information processed: Confirmed. Only de-identified sample data is used + +2. Open source in use + - SBOM: fds-research-env.sbom.json (187 Python-ecosystem components) + - Risk assessment: 0 critical vulnerabilities, 0 copyleft licenses (all Apache-2.0, BSD, MIT) + +3. Additional risks and controls + - Additional risk: inflow of unverified components through direct access to external package repositories + - Alternative information security controls: endpoint controls and access logs according to the criteria for alternative information security controls in place of network separation + - Security controls: internal mirror used first, weekly vulnerability scan of the entire environment + +4. Implementation check and reassessment + - Implementation check: quarterly review of control operation (information security department) + - Reassessment cycle: once a year and whenever the environment configuration changes + +5. Review and approval + - Author: Park OO (FDS Research Team) / 2026-04-02 + - Review: information security department (2026-04-09), open source management organization (SBOM and risk assessment confirmed) + - Approval: Information Security Committee / 2026-04-16 +``` + +This report becomes the basis for approval in [Usage +Approval](../../4-approve/#self-risk-assessment-for-network-separation-exceptions) and is updated +periodically in [Management](../../5-manage/#periodic-reassessment). + +--- + +*Last reviewed: 2026-06-10. This page is reviewed when regulations change and once a year on a regular basis.* diff --git a/content/en/guide/finance-oss-guide/artifacts/3-audit-evidence/_index.md b/content/en/guide/finance-oss-guide/artifacts/3-audit-evidence/_index.md new file mode 100644 index 0000000000..82c0b7caa5 --- /dev/null +++ b/content/en/guide/finance-oss-guide/artifacts/3-audit-evidence/_index.md @@ -0,0 +1,83 @@ +--- +title: "Audit Evidence List" +linkTitle: "Audit Evidence List" +weight: 30 +type: docs +categories: ["guide"] +tags: ["Finance", "Open Source", "Audit", "Evidence"] +description: > + Organizes, as a checklist, the open source management evidence required in internal and external + audits and in information technology inspections by the Financial Supervisory Service, and + specifies which activity produces each item and where it is kept. +--- + +The financial sector must keep evidence of its open source management activities to prepare for +internal and external audits and for information technology (IT) inspections by the Financial +Supervisory Service (금융감독원). Because check records from the management step serve directly as +evidence, do not create it separately; collect the records that come out of everyday activities in +a systematic way. For more context, see [Management](../../5-manage/#audit-evidence-management). + +## Evidence checklist + +The evidence below overlaps with the verification materials of ISO/IEC 5230 and 18974. Documents +created while preparing for ISO self-certification serve directly as audit evidence. Set the +storage location and retention period according to your organization's rules. + +| Evidence | Producing activity | Related ISO verification material | Storage location (example) | Notes | +|------|------|------|------|------| +| Open source policy document and revision history | Governance | 5230 3.1.1.1 | Document management system | Including version history | +| Role and responsibility documents, records of assigned owners | Governance | 5230 3.1.2.1, 3.2.2.1 | Document management system | | +| SBOMs and update history | Identification | 5230 3.3.1.1, 3.3.1.2 | SBOM management tool | Per system | +| Vulnerability check results | Identifying and resolving issues | 18974 4.3.2.1, 4.3.2.2 | Vulnerability management tool | By date | +| Vulnerability remediation records (including decisions that no action is needed) | Identifying and resolving issues | 18974 4.3.2.2 | Vulnerability management tool | Including VEX | +| License review records | Identifying and resolving issues | 5230 3.3.2.1 | Document management system | | +| Usage approval request, review, and decision records | Usage approval | 5230 3.1.5.1 | Approval management tool | Basis for board decisions | +| Self risk assessment report for network separation exceptions | Usage approval | (Regulation on Supervision of Electronic Financial Transactions (전자금융감독규정)) | Document management system | Information Security Committee (CISO) approval history, reassessment history | +| Import approval records | Closed network operations | - | Document management system | Including verification results | +| Periodic reassessment results | Management | 18974 4.1.2.5 | Document management system | Per cycle | +| Compliance artifacts (notices, etc.) | Management | 5230 3.4.1.1, 3.4.1.2 | Distribution artifact repository | Distributed software | +| Open source requirement clauses in outsourcing contracts and submitted SBOMs | Usage approval | (Internal control under Article 21 of the Regulation on Supervision of Electronic Financial Transactions) | Contract management system | Electronic financial auxiliary service providers (전자금융보조업자) | + +## Retention and management principles + +Protecting evidence is as important as creating it. + +- Set the retention period. Decide it with the audit cycle and regulatory requirements in mind, + confirm that it does not conflict with retention periods set by regulations, such as the retention + period for electronic financial transaction records, and then state it in the artifact retention + procedure (ISO/IEC 5230 3.4.1.2). +- Prevent tampering. Keep records in a way that cannot be altered later (append-only logs, access + control). +- Make records traceable. Record the actor and the time so that you can tell who did what and when. +- Check inspection readiness in advance. At periodic reassessments, also confirm that all evidence is + being kept. + +{{% alert title="Field case: KakaoBank's audit response" color="success" %}} +At the 30th KWG regular meeting (2026-06), KakaoBank shared its experience responding to the +Financial Supervisory Service's IT risk quantitative assessment and management condition assessment, +the Bank of Korea (한국은행)'s annual survey on the status of financial informatization, and data +requests from the Korea Deposit Insurance Corporation (예금보험공사). What the institutions commonly +ask about comes down to five things: whether the open source inventory is managed, the full +inventory list, the composition of management staff, policies and internal rules, and the security +vulnerability check procedure. These mostly overlap with the evidence checklist above. + +![Presentation slide summarizing the five things institutions commonly check in inspections and data requests](./30th-s3-key-questions.png) + +*Figure: The five things institutions commonly check. Minae Lee (KakaoBank), slides, page 5. The +slide image is the presenter's work and is not covered by the CC BY 4.0 license that applies to this +text.* + +For an inquiry about the purpose of and reason for using components that are no longer supported +(deprecated), KakaoBank could answer with the management records it had kept as a matter of course +(purpose of use, confirmation that there were no security vulnerabilities). The case shows that a +record showing the organization knows about and manages old components matters more than whether +such components exist at all. However, because each institution requires a different scope and +format for the list, converting the SBOM into the requested format remains necessary even when SBOMs +are in place. + +Source: Minae Lee (KakaoBank), "Experience Report on Open Source Operations as a Financial Company" (금융회사로서의 오픈소스 관련 업무 대응 후기), [KWG 30th meeting (2026-06) slides](https://github.com/OpenChain-Project/OpenChain-KWG/releases/download/meeting-slides-2026/30th-session3-finance-oss-report.pdf). +{{% /alert %}} + +--- + +*Last reviewed: 2026-06-10. This page is reviewed when regulations change and once a year on a regular basis.* diff --git a/content/en/guide/finance-oss-guide/artifacts/4-tool-recipe/_index.md b/content/en/guide/finance-oss-guide/artifacts/4-tool-recipe/_index.md new file mode 100644 index 0000000000..f8995243cf --- /dev/null +++ b/content/en/guide/finance-oss-guide/artifacts/4-tool-recipe/_index.md @@ -0,0 +1,143 @@ +--- +title: "Tool Setup Recipe" +linkTitle: "Tool Setup Recipe" +weight: 40 +type: docs +categories: ["guide"] +tags: ["Finance", "Open Source", "Tools", "Dependency-Track", "Closed Network"] +description: > + A docker-compose example and integration procedure for setting up SBOM generation and continuous + vulnerability monitoring in an on-premises closed network environment. It shows the flow of + creating SBOMs with cdxgen and registering them in Dependency-Track. +--- + +This recipe is an example of setting up SBOM (Software Bill of Materials) generation and continuous +vulnerability monitoring in a closed network. As a quick starting point, it reorganizes the existing +[cdxgen and Dependency-Track pipeline tutorial](../../../tools/8-cdxgen-dt/) for the financial sector. +The purpose is not to recommend a specific product but to show how to build the setup with open +source, on-premises tools. + +{{% alert title="Closed network assumptions" color="warning" %}} +The example below includes a step that downloads tool images. In a closed network, obtain the images +and vulnerability databases from outside, move them to the internal container registry and mirror, +and configure the tools to pull only from inside. For the import procedure, see [Closed Network +Operation](../../0-closed-network/#import-control). +{{% /alert %}} + +## Setting up Dependency-Track + +When SBOMs are registered in [Dependency-Track](../../../tools/7-dependency-track/), it automatically +reassesses the affected components each time the vulnerability database is updated. Use it for +continuous monitoring of operational systems. Below is a docker-compose example that runs the API +server and the frontend together. + +```yaml +# docker-compose.yml +volumes: + dependency-track: + +services: + dtrack-apiserver: + image: dependencytrack/apiserver:latest + deploy: + resources: + limits: + memory: 12288m + reservations: + memory: 8192m + ports: + - "8081:8080" + volumes: + - "dependency-track:/data" + restart: unless-stopped + + dtrack-frontend: + image: dependencytrack/frontend:latest + depends_on: + - dtrack-apiserver + environment: + - API_BASE_URL=http://localhost:8081 + ports: + - "8080:8080" + restart: unless-stopped +``` + +In a closed network, change the `image` values to internal registry addresses. For example, instead +of `dependencytrack/apiserver:latest`, point to an image mirrored internally, such as +`registry.internal/dependencytrack/apiserver:4.x`. For the version tag, it is safer to use a +verified fixed version instead of `latest`. + +`API_BASE_URL` is the address the user's browser uses to call the API server, so when deploying on +an internal server, replace `localhost` with that server's host name (for example, +`http://dtrack.internal:8081`). + +After starting the services, log in, change the initial administrator password, and issue an API key +for uploading SBOMs. + +```bash +# Start the services +docker compose up -d + +# Check the status +docker compose ps +``` + +Depending on the environment, you may need to use the `docker-compose` (hyphenated) command instead +of `docker compose` (plugin). The hyphenated command (legacy v1) ignores the `deploy.resources` +memory limits in the example, so add the `--compatibility` option to apply the limits. + +![Dependency-Track dashboard after setup](./dt-dashboard.png) + +*Figure: The Dependency-Track dashboard you see after finishing the setup and registering SBOMs. The +screen reuses a capture from the [cdxgen and Dependency-Track tutorial](../../../tools/8-cdxgen-dt/). +Step-by-step screens for account setup and API key issuance are also in that tutorial.* + +## Integrating SBOM generation and registration + +Create the project's SBOM with [cdxgen](../../../tools/5-cdxgen/) and upload the result to Dependency-Track. The uploaded SBOM +is then reassessed automatically each time the vulnerability database is updated. cdxgen accesses +build tools and package repositories while resolving dependencies for each language ecosystem, so in +a closed network, run it with the repository settings for npm, Maven, and so on pointing to the +internal mirror. + +```bash +# 1) Generate a CycloneDX SBOM with cdxgen +# -r recursively collects multiple manifests in subdirectories +cdxgen --spec-version 1.6 -r -o sbom.json /path/to/project + +# 2) Upload the SBOM to a Dependency-Track project +curl -X POST "http://localhost:8081/api/v1/bom" \ + -H "X-Api-Key: ${DT_API_KEY}" \ + -F "projectName=my-service" \ + -F "projectVersion=1.0.0" \ + -F "autoCreate=true" \ + -F "bom=@sbom.json" +``` + +The latest version of cdxgen generates CycloneDX 1.7 by default, but depending on the +Dependency-Track version, only up to 1.6 may be accepted. If the upload is rejected, specify +`--spec-version 1.6` as shown above to match the format. `autoCreate=true` creates a new project if +none exists with the same name and version. If you keep a project for each operational system and +register its SBOM, you can see in one place which systems are affected when a new vulnerability is +disclosed. This flow implements the continuous monitoring covered in +[Management](../../5-manage/#continuous-vulnerability-monitoring). + +## Offline vulnerability database + +In a closed network, configure the addresses of the vulnerability data sources that Dependency-Track +refers to so that they point to internal mirrors, using the administration screen. The difficulty of +setting up a mirror for each data source and the preliminary verification it needs, as well as how +cache files are imported for command-line scanning tools (Grype, Trivy), are covered in +[Offline vulnerability management in Closed Network Operation](../../0-closed-network/#offline-vulnerability-management). + +## Tool selection criteria + +This recipe uses cdxgen and Dependency-Track as examples, but you can replace them with other open +source tools that do the same job. When choosing tools, follow the criteria presented in [Closed +Network Operation](../../0-closed-network/#choosing-tools-for-a-closed-network) (on-premises +installation, offline database updates, standard-format input and output, and the license of the +tool itself). Installation and usage for each tool are described on the [tools page](../../../tools/). + +--- + +*Last reviewed: 2026-06-10. This page is reviewed when regulations change and once a year on a regular basis.* diff --git a/content/en/guide/finance-oss-guide/artifacts/_index.md b/content/en/guide/finance-oss-guide/artifacts/_index.md new file mode 100644 index 0000000000..0060cbcf02 --- /dev/null +++ b/content/en/guide/finance-oss-guide/artifacts/_index.md @@ -0,0 +1,33 @@ +--- +title: "Artifacts: Ready-to-Use Forms and Recipes" +linkTitle: "Artifacts" +weight: 80 +type: docs +categories: ["guide"] +tags: ["Finance", "Open Source", "Template", "Workbook", "Audit"] +description: > + Collects the four practical artifacts provided with the guide text: the self-check workbook, + policy and procedure templates, the audit evidence list, and the tool setup recipe. +--- + +This guide provides four ready-to-use artifacts along with the main text. Where the main text +explains "what to do and why," the artifacts fill in "in what form to record it." Each artifact is +meant to be adapted to your organization's situation. + +| Artifact | Use | +|------|------| +| [Self-Check Workbook](./1-workbook/) | Combines check items, ISO verification materials, and compliance status, owner, and deadline in a single sheet to find the gaps in your program | +| [Policy and Procedure Templates](./2-policy-templates/) | Provides outlines for a finance-specific policy, an import procedure, a usage approval form, and a self risk assessment report for network separation exceptions | +| [Audit Evidence List](./3-audit-evidence/) | Organizes, as a checklist, the evidence required in internal and external audits and supervisory inspections and where it is kept | +| [Tool Setup Recipe](./4-tool-recipe/) | Provides an example of setting up and integrating on-premises SBOM and vulnerability tools in a closed network with docker-compose | + +{{% alert title="How to use" color="info" %}} +The artifacts supplement the existing [policy and procedure templates](../../templates/) guide for +the financial context. Use the existing templates for general practice forms, and use these +artifacts together with them for items specific to the financial sector (import control, risk +assessment for network separation exceptions, and SBOM requirements for outsourcing). +{{% /alert %}} + +--- + +*Last reviewed: 2026-06-10. This page is reviewed when regulations change and once a year on a regular basis.* diff --git a/content/en/guide/iso42001_guide/1-context-leadership/_index.md b/content/en/guide/iso42001_guide/1-context-leadership/_index.md new file mode 100644 index 0000000000..aa36575453 --- /dev/null +++ b/content/en/guide/iso42001_guide/1-context-leadership/_index.md @@ -0,0 +1,148 @@ +--- +title: "1. Context of the Organization and Leadership" +weight: 10 +type: docs +categories: ["guide"] +tags: ["ISO/IEC 42001", "AI Policy", "Context of the Organization", "Leadership"] +description: > + Explains the requirements of ISO/IEC 42001 §4 (Context of the organization) and §5 (Leadership) that intersect with open source management. + It describes how to include open source usage principles in the AI policy. +--- + +## 1. Overview + +ISO/IEC 42001 §4 and §5 are the **foundation** of the AI management system. They determine which AI systems the organization operates, +who the interested parties are, and on what principles top management will manage AI. + +From the perspective of open source practitioners, §5.2 (AI policy) is the most important point of intersection. +Including principles on open source use in the AI policy makes it possible to operate it in an integrated way with the ISO 5230 policy. + +--- + +## 2. §4 Context of the Organization: Open Source Considerations + +### §4.1 Understanding the Organization and Its Context + +When identifying internal and external issues that affect the purpose of the AI management system, +include external issues related to open source: + +- Changes in the license policies of open source AI models (for example, changes to Llama license terms) +- Domestic and international AI regulations and mandatory open source disclosure requirements +- Security vulnerability risks of open source components in the AI supply chain + +#### Global AI Regulations and Standards Matrix (as of 2026-05) + +The following global regulations and standards directly intersect with open source management in AI systems, +so they must be reflected in the analysis of the organization's context. This matrix is the reference baseline for this entire guide. + +| Regulation or standard | Effective date | Key obligations | Open source intersection | +|----------|-----------|----------|---------------| +| **EU AI Act §53** (GPAI) | Phased application from 2026-08 | Publishing a summary of training data, respecting copyright opt-outs, preparing technical documentation | Tracking licenses and sources of AI model training datasets | +| **EU AI Act §50** | 2026-08 | Labeling and transparency for AI-generated content | Marking the output of AI coding tools | +| **EU AI Act §40** | 2026-08 | Reporting energy consumption | AI SBOM `energyConsumption` field | +| **EU AI Act §25** | 2026-08 | Value chain obligations (providing information to downstream integrators) | Receiving supplier information when adopting external AI models | +| **EU AI Act §11 + Annex IV** | 2027 (high-risk) | Obligation to prepare technical documentation | AI SBOM as a core element of technical documentation | +| **Korea's AI Basic Act (AI 기본법)** | In force from 2026-01 | Classification of high-impact AI, impact assessment, AI system labeling obligations, disclosure of training data sources | Obligation to prepare AI SBOMs and model cards | +| **US Copyright Office AI guidance** | Published 2023-03, updated 2024 | Fully AI-generated works = no human authorship (cannot be registered for company copyright). Copyright is recognized according to the proportion of human review and modification | Handling copyright ownership of AI coding tool output | +| **US EO 14110** (2023) | Revoked 2025-01 (replaced by EO 14179) | SBOM requirements for federal AI systems, reporting of dual-use AI models | AI SBOM standardization, adoption of NIST AI RMF | +| **NIST AI RMF 2.0** | Published 2024 | Standard framework for AI risk management (GOVERN, MAP, MEASURE, MANAGE) | Complementary to ISO/IEC 42001 | +| **OSAID 1.0** (OSI, 2024-10) | In effect | Definition of an "open source AI model": disclosure of three elements (data, code, weights) | "Open Weight" models (Llama, Gemma) do not meet OSAID, so classification in tables must be accurate | +| **ISO/IEC 42005** (2025) | Published | Standard for AI system impact assessment | Specific techniques for the §6.1.4 impact assessment | +| **ISO/IEC 23894** (2023) | Published | Standard for AI risk management | Supplements the §6.1.2 risk assessment | +| **ISO/IEC 5338** (2023) | Published | Standard for AI system life cycle | Defines life cycle stages for §8.5 | +| **ISO/IEC 42003** | In development (new project stage) | Implementation guidance for AI management systems | OpenChain AI WG is working to provide input | +| **ISO/IEC 42006** | Published (2025-07) | Requirements for AI certification bodies | Based on ISO 17021-1, reliability of third-party certification | + +{{% alert title="Using the matrix in context analysis" color="info" %}} +Map this matrix to external issues when analyzing the organization's context (§4.1). The other pages of this guide +(§6.1.2 risk assessment, §6.1.4 impact assessment, §7.5 documentation, §8.5 system life cycle, +§8.6 data, §8.8 external procurement) also refer to this matrix. + +The regulations and standards listed in the matrix are as of 2026-05. Update this table +when a new version is published or an effective date changes. +{{% /alert %}} + +### §4.3 Determining the Scope of the AI Management System + +When defining the scope of the AI management system, explicitly include AI systems and processes +that use open source. + +**Checkpoints**: +- [ ] Does the AI management system scope document include AI systems that use open source frameworks, models, and datasets? +- [ ] Are external issues related to open source licenses included in the analysis of the organization's context? + +--- + +## 3. §5.2 AI Policy: Including Open Source Principles ★ + +ISO/IEC 42001 requires the organization to establish an **AI policy**. +From an open source compliance perspective, including the following in the AI policy +keeps it consistent with the ISO 5230 policy. + +### Open Source Principles to Add to the AI Policy + +```markdown +## Principles for Open Source Use in AI Systems + +This organization complies with the following principles when using open source in the development and operation of AI systems. + +1. **License compliance** + Review and fulfill the license terms of all open source frameworks, pre-trained models, + and training datasets used in AI systems. + +2. **AI SBOM management** + Document the open source components (frameworks, models, datasets) that make up AI systems + in an AI SBOM and keep it up to date. + +3. **Security vulnerability management** + Regularly check the open source components used in AI systems for security vulnerabilities + and respond promptly to identified vulnerabilities. + +4. **Supply chain verification** + Confirm the components and licenses of open source AI models and AI services procured externally. +``` + +### Integration with the Existing Open Source Policy + +{{% alert title="How to integrate with the ISO/IEC 5230 policy" color="success" %}} + +If the company already operates an open source policy based on ISO/IEC 5230, +it can operate them in an integrated way by **adding an AI section to the existing open source policy** +instead of writing separate open source principles for the AI policy. + +- How to establish an ISO 5230 policy: [Enterprise Open Source Management Guide: 2. Policy](../../opensource_for_enterprise/2-policy/) +- Open source policy template: [Templates: Open Source Policy](../../templates/1-policy/) + +{{% /alert %}} + +**Checkpoints**: +- [ ] Does the AI policy or the open source policy include principles for open source use in AI systems? +- [ ] Has the AI policy been approved by top management and communicated within the organization? +- [ ] Are there no conflicts between the open source policy and the AI policy? + +--- + +## 4. §5.3 Roles, Responsibilities, and Authorities + +Clarify roles and responsibilities related to open source in the AI management system. + +| Role | ISO 42001 responsibility | Additional open source responsibility | +|------|---------------|----------------------| +| AI governance lead | Establishing and overseeing the AI policy | Ensuring OSS principles are included in the AI policy | +| AI development lead | Managing the AI system life cycle | Creating and maintaining AI SBOMs | +| Open source program manager (OSPM) | - | Overall OSS compliance of AI systems | +| Security owner | AI risk management | Monitoring AI OSS vulnerabilities | + +{{% alert title="Recommendation" color="success" %}} +Formalizing the role so that the open source program manager (OSPM) participates in the AI governance committee +or AI review process allows compliance in both areas to be integrated efficiently. +{{% /alert %}} + +--- + +## 5. References + +- [ISO/IEC 42001 Guide home](../) +- [Standards Comparison: ISO 5230, 18974, 42001](../compare/) +- [Enterprise Open Source Management Guide: 2. Policy](../../opensource_for_enterprise/2-policy/) +- [Enterprise Open Source Management Guide: 1. Organization](../../opensource_for_enterprise/1-teams/) diff --git a/content/en/guide/iso42001_guide/2-planning/_index.md b/content/en/guide/iso42001_guide/2-planning/_index.md new file mode 100644 index 0000000000..7ab9cdedb6 --- /dev/null +++ b/content/en/guide/iso42001_guide/2-planning/_index.md @@ -0,0 +1,154 @@ +--- +title: "2. Planning" +weight: 20 +type: docs +categories: ["guide"] +tags: ["ISO/IEC 42001", "AI Risk", "Impact Assessment", "Planning"] +description: > + Explains the requirements of ISO/IEC 42001 §6 (Planning) that intersect with open source management. + It describes how to include open source license and vulnerability risks in the AI risk assessment and the AI system impact assessment. +--- + +## 1. Overview + +ISO/IEC 42001 §6 requires a systematic way to identify and address risks and opportunities related to AI systems. +In particular, §6.1.2 (AI risk assessment) and §6.1.4 (AI system impact assessment) are +used directly to include risks that can arise from open source components. + +--- + +## 2. §6.1.2 AI Risk Assessment: Including Open Source Risks ★ + +ISO/IEC 42001 requires a process to identify, analyze, and evaluate risks related to AI systems. +Open source practitioners must explicitly include **open source license risks** and **security vulnerability risks** +in this process. + +### Types of Open Source-Related AI Risks + +| Risk type | Cause | Impact | Response | +|-----------|------|------|---------| +| **License non-compliance** | Custom license terms of an AI model not checked | Legal disputes, service suspension | Mandatory license review before use | +| **GPL contamination** | Combining copyleft-licensed AI components with proprietary code | Obligation to disclose source code | Prior review of license compatibility | +| **Violation of model license terms** | MAU limits or use restrictions not met | License termination, damages | Regular checks of compliance with each term | +| **Open source vulnerabilities** | CVEs in AI frameworks or their dependencies | System compromise, data leakage | Regular SCA scans, applying patches | +| **Dataset license violation** | Using CC-BY-NC data in a commercial service | Copyright infringement | Re-review of dataset licenses | +| **Supply chain risk** | License change of an external open source model | Threat to service continuity | Establishing supply chain monitoring | + +### Open Source Items to Include in the AI Risk Assessment Report + +```markdown +## AI Risk Assessment Report: Open Source Risk Section + +### AI system assessed: [System name] +### Assessment date: YYYY-MM-DD + +| # | Risk item | Component currently used | Likelihood | Impact | Risk level | Response | +|---|------------|-----------------|:----:|:----:|:---------:|---------| +| 1 | License non-compliance | [Model name, license] | Medium | High | High | Legal review | +| 2 | OSS vulnerability | [Framework name, version] | Medium | High | High | SCA scan | +| 3 | Dataset license | [Dataset name, license] | Low | Medium | Medium | Recheck license | +``` + +{{% alert title="Connection with ISO/IEC 18974" color="success" %}} +The risk of open source vulnerabilities used in AI systems connects directly to the §4.3.2 (Security assurance) requirement +of ISO/IEC 18974. Applying the vulnerability detection and assessment process of 18974 to AI systems as well removes the need to build +a separate risk assessment framework. + +Related guide: [ISO/IEC 18974: §4.3.2 Security Assurance](../../iso18974_guide/3-content-review/2-security-assurance/) +{{% /alert %}} + +**Checkpoints**: +- [ ] Does the AI risk assessment process include open source license risk items? +- [ ] Does the AI risk assessment process include open source security vulnerability risk items? +- [ ] Are the results of the risk assessment documented? +- [ ] Has a treatment plan been established according to the risk level? + +--- + +## 3. §6.1.4 AI System Impact Assessment: Using ISO/IEC 42005 ★ + +When assessing the impact an AI system can have on individuals, groups, and society, +also include the impact caused by open source components. + +{{% alert title="ISO/IEC 42005:2025, AI System Impact Assessment" color="info" %}} +ISO/IEC 42005:2025 (published 2025-05) is an SC 42 family standard that provides the **specific method of performing** +the §6.1.4 AI system impact assessment. ISO 42001 states only the principle that "an impact assessment shall be performed," +but 42005 provides guidance on the assessment scope, timing, interested parties, and even the documentation format. + +**Recommended use**: ISO 42001 certification auditors use the eight assessment areas of 42005 as the criteria +when checking how specific the impact assessment deliverables are. Organizations that comply only with 42001 without referring to 42005 +are likely to be told that their impact assessment reports are a formality. +{{% /alert %}} + +### 3.1 Eight Impact Assessment Areas Based on ISO/IEC 42005 + +| # | Assessment area | Key question (open source perspective) | +|---|----------|------------------------| +| 1 | **Affected individuals and groups** | Which population groups are affected in a discriminatory way by known biases of the open source model? | +| 2 | **Societal and environmental impact** | Are the training data sources biased toward particular languages or cultures? What are the carbon emissions of GPU training? | +| 3 | **Intended use cases** | Does the open source model license permit the intended use (for example, the military and medical exclusion clauses of Llama)? | +| 4 | **Reasonably foreseeable misuse** | What misuse paths are possible because the model weights are public (deepfakes, generating weapons information, and so on)? | +| 5 | **Data quality and representativeness** | Bias, licenses, and inclusion of personal information in the training dataset (open data) | +| 6 | **System limitations and failure modes** | Are the limitations stated in the model card sufficiently communicated to users? | +| 7 | **Explainability and transparency** | Level of disclosure of the AI SBOM (SPDX 3.0/CycloneDX 1.6 ML-BOM) | +| 8 | **Governance and accountability** | Allocation of responsibility when the open source model has defects (including limits on supplier indemnification) | + +### 3.2 AI System Impact Assessment Report Template + +```markdown +## AI System Impact Assessment Report (ISO/IEC 42001 §6.1.4 + ISO/IEC 42005) + +### 1. System overview +- System name / version / assessment date / assessor +- Intended use cases / out-of-scope use cases +- Open source models, datasets, and frameworks used (see the AI SBOM) + +### 2. Identifying interested parties +- Direct users / indirectly affected groups / regulators / data subjects + +### 3. Impact analysis by area (eight areas of 42005) +- Area 1 (affected individuals and groups): __________ +- Area 2 (societal and environmental): __________ +- ... (complete all eight areas) + +### 4. Impacts related to open source components +- Impact of security vulnerabilities: __________ +- Impact of service suspension due to copyright issues: __________ +- Transparency of training data sources: __________ + +### 5. Mitigation measures +- Mitigation measure / owner / due date for each identified risk + +### 6. Assessment result and approval +- Residual risk level / deployment approval / next reassessment date +``` + +**Checkpoints**: +- [ ] Do the AI system impact assessment items include security and legal impacts related to open source components? +- [ ] Is the impact assessment performed before deploying the AI system and when major changes occur? +- [ ] Does it cover all eight assessment areas of ISO/IEC 42005? +- [ ] Are the assessment results appropriately disclosed to interested parties (including mitigation measures)? + +--- + +## 4. §6.2 AI Objectives + +Establish measurable AI objectives linked to open source compliance. + +### Examples of Open Source-Related AI Objectives + +```markdown +- AI SBOM currency rate: maintain 100% each quarter +- Open source vulnerability patch completion rate: 95% or more of Critical CVEs patched within 14 days of discovery +- AI model license review completion rate: 100% prior review when adopting new models +- Dataset license recording rate: 100% of dataset licenses recorded in the AI SBOM +``` + +--- + +## 5. References + +- [ISO/IEC 42001 Guide home](../) +- [ISO/IEC 18974: §4.3.2 Security Assurance](../../iso18974_guide/3-content-review/2-security-assurance/) +- [Enterprise Open Source Management Guide: 3. Process](../../opensource_for_enterprise/3-process/) +- [ISO 5230, 18974, 42001 Comparison: SC 42 Family Mapping](../compare/#3-sc-42-family-mapping) diff --git a/content/en/guide/iso42001_guide/3-support/_index.md b/content/en/guide/iso42001_guide/3-support/_index.md new file mode 100644 index 0000000000..bd125823a4 --- /dev/null +++ b/content/en/guide/iso42001_guide/3-support/_index.md @@ -0,0 +1,145 @@ +--- +title: "3. Support" +weight: 30 +type: docs +categories: ["guide"] +tags: ["ISO/IEC 42001", "Competence", "AI SBOM", "Documentation", "Support"] +description: > + Explains the requirements of ISO/IEC 42001 §7 (Support) that intersect with open source management. + It describes how to establish competence in AI open source management and how to document the AI SBOM. +--- + +## 1. Overview + +ISO/IEC 42001 §7 covers the **competence, awareness, communication, and +documentation** needed to operate the AI management system. From the perspective of open source practitioners, **§7.2 (Competence)** and §7.5 (Documented information) are +the key points of intersection. + +--- + +## 2. §7.2 Competence: AI Open Source Management Competence ★ + +ISO/IEC 42001 requires that persons performing AI-related roles have the necessary competence. +From an open source compliance perspective, AI development and operations staff should have the following competencies. + +### Elements of AI Open Source Management Competence + +| Competence area | Description | Related training | +|---------|------|---------| +| **Understanding AI framework licenses** | Knowing the license terms of major frameworks such as PyTorch and TensorFlow | Open source license training | +| **Reviewing AI model licenses** | Analyzing custom license terms (use restrictions, handling of derivatives) | Training on collaboration with the legal team | +| **Writing AI SBOMs** | How to build an AI SBOM based on the SPDX 3.0 AI profile | AI SBOM hands-on practice | +| **Open source vulnerability management** | SCA scanning of AI system dependencies and CVE response | DevSecOps training | +| **Open data licenses** | CC family license terms and rules for using datasets | Data license training | + +### Integration with the Existing Open Source Competence Framework + +{{% alert title="Extending the ISO/IEC 5230 competence framework" color="success" %}} +If the company already operates a competence and training framework based on ISO/IEC 5230, +extend it by adding **AI-specific modules** to the existing open source training courses. + +Examples of additional modules: +- Understanding AI model licenses (1 hour) +- AI SBOM hands-on practice (2 hours) +- How to verify open source in the AI supply chain (1 hour) + +Related guide: [Enterprise Open Source Management Guide: 5. Training](../../opensource_for_enterprise/5-training/) +{{% /alert %}} + +**Checkpoints**: +- [ ] Do AI development staff understand the license terms of AI frameworks and models? +- [ ] Is training on how to write AI SBOMs provided? +- [ ] Are competence evaluation records documented? + +--- + +## 3. §7.5 Documented Information: AI SBOM ★ + +ISO/IEC 42001 §7.5 requires establishing and maintaining the **documented information** needed to operate the AI management system. +From an open source perspective, the **AI SBOM (AI System Bill of Materials)** is the key deliverable of §7.5. + +### What is an AI SBOM? + +An AI SBOM extends the software SBOM (§3.3.1 of ISO 5230) to AI systems. +It is a list that documents every element that makes up an AI system, along with its source and license. + +| Category | Conventional software SBOM | AI SBOM | +|------|---------------------|---------| +| Scope | Libraries, packages | Frameworks + models + datasets | +| Licenses | SPDX standard licenses | SPDX + custom AI licenses | +| Standard format | SPDX 2.x, CycloneDX | SPDX 3.0 AI profile | +| Additional metadata | None | Number of model parameters, training data sources, and so on | + +### Components of an AI SBOM + +``` +AI SBOM + ├── 1. AI frameworks and libraries + │ name, version, license, hash + │ + ├── 2. Pre-trained Model + │ name, version, license, modelCard URL + │ primaryPurpose (inference / training / fine-tuning) + │ + ├── 3. Training datasets + │ name, version, license, datasetType + │ knownBias (bias information, if any) + │ + └── 4. Fine-tuning data (if applicable) + name, license, source +``` + +Details on writing an AI SBOM: [AI SBOM Guide](../4-operation/2-ai-sbom/) + +### Brief Example of the SPDX 3.0 AI Profile + +```yaml +SPDXVersion: SPDX-3.0 +DataLicense: CC0-1.0 + +# AI framework +- SPDXID: SPDXRef-pytorch + name: pytorch + versionInfo: "2.2.0" + licenseConcluded: BSD-3-Clause + primaryPurpose: library + +# Pre-trained model +- SPDXID: SPDXRef-llama3 + name: meta-llama/Llama-3.1-8B + versionInfo: "3.1" + licenseConcluded: LicenseRef-Llama-Community + primaryPurpose: inference + modelCard: "https://huggingface.co/meta-llama/Llama-3.1-8B" + +# Training dataset (for fine-tuning) +- SPDXID: SPDXRef-dataset-alpaca + name: tatsu-lab/alpaca + licenseConcluded: CC-BY-4.0 + dataCollectionProcess: "Stanford Alpaca 52K instruction dataset" +``` + +**Checkpoints**: +- [ ] Are all frameworks, models, and datasets used in the AI system recorded in the AI SBOM? +- [ ] Is the AI SBOM updated every time the AI system changes? +- [ ] Does the AI SBOM include license information for each component? + +--- + +## 4. §7.3 Awareness + +Ensure that persons performing roles related to AI systems are aware of the importance of open source compliance. + +**Checkpoints**: +- [ ] Are AI development staff aware of the risks of not complying with open source requirements in AI systems? + +--- + +## 5. References + +- [ISO/IEC 42001 Guide home](../) +- [AI SBOM Guide](../4-operation/2-ai-sbom/) +- [ISO/IEC 5230: §3.3.1 SBOM](../../iso5230_guide/3-content-review/1-sbom/) +- [ISO/IEC 18974: §4.3.1 SBOM](../../iso18974_guide/3-content-review/1-sbom/) +- [Enterprise Open Source Management Guide: 5. Training](../../opensource_for_enterprise/5-training/) +- [AI Work Group](../../../resource/AI_work_group/) diff --git a/content/en/guide/iso42001_guide/4-operation/1-oss-in-ai/_index.md b/content/en/guide/iso42001_guide/4-operation/1-oss-in-ai/_index.md new file mode 100644 index 0000000000..178ff45c0e --- /dev/null +++ b/content/en/guide/iso42001_guide/4-operation/1-oss-in-ai/_index.md @@ -0,0 +1,155 @@ +--- +title: "Open Source Management in AI Systems" +weight: 10 +type: docs +categories: ["guide"] +tags: ["ISO/IEC 42001", "AI Framework", "Open Source License", "AI Data"] +description: > + Explains how to manage the open source licenses of AI frameworks, pre-trained models, and training datasets + in accordance with ISO/IEC 42001 §8.5 (AI system life cycle) and §8.6 (AI data management). +--- + +## 1. Overview + +ISO/IEC 42001 §8.5 (AI system life cycle) and §8.6 (AI data management) cover requirements that must be managed +during the AI system development stage. From an open source perspective, the key is +**license compliance for the open source components** used in AI systems. + +--- + +## 2. §8.5 AI System Life Cycle: Managing Open Source Frameworks and Models + +### 2.1 Planning and Design Stage + +When selecting the open source frameworks and models to use in the planning stage of an AI system, +review their licenses in advance. + +**Prior license review procedure:** + +``` +1. List candidate components + (framework name, version, source URL) + +2. Check licenses + - Frameworks: check PyPI, npm, GitHub README + - AI models: check the Hugging Face Model Card and official repository + +3. License terms to review + □ Whether commercial use is permitted + □ Whether there is an obligation to disclose derivatives (fine-tuned models) + □ Whether there is a copyright notice obligation + □ Whether there is a patent clause (such as the explicit patent grant in Apache 2.0) + □ Whether there are use restrictions based on MAU or revenue + +4. If an unacceptable license is identified, look for alternative components +``` + +### 2.2 Development Stage + +Each time a new open source component is added during development, check its license +and reflect it in the AI SBOM immediately. + +**Using SCA (software composition analysis) tools:** + +Apply existing open source scanning tools to AI project repositories in the same way. + +```bash +# Example of scanning an AI project with FOSSLight +fosslight -p ./ai-project-dir -o ./ai-sbom-output + +# Generate an SBOM for a Python AI project with cdxgen +cdxgen -t python ./ai-project-dir -o ai-sbom.json +``` + +Related tool guides: +- [FOSSLight](../../../tools/3-fosslight/) +- [cdxgen](../../../tools/5-cdxgen/) + +### 2.3 Details of Major AI Framework Licenses + +| Framework | License | Copyright notice | Patent grant | Commercial use | Source disclosure | +|-----------|---------|:---------:|:-------:|:-------:|:-------:| +| PyTorch | BSD 3-Clause | ✅ Required | ❌ None | ✅ Allowed | ❌ Not required | +| TensorFlow | Apache 2.0 | ✅ Required | ✅ Yes | ✅ Allowed | ❌ Not required | +| JAX | Apache 2.0 | ✅ Required | ✅ Yes | ✅ Allowed | ❌ Not required | +| Hugging Face Transformers | Apache 2.0 | ✅ Required | ✅ Yes | ✅ Allowed | ❌ Not required | +| LangChain | MIT | ✅ Required | ❌ None | ✅ Allowed | ❌ Not required | +| scikit-learn | BSD 3-Clause | ✅ Required | ❌ None | ✅ Allowed | ❌ Not required | +| ONNX Runtime | MIT | ✅ Required | ❌ None | ✅ Allowed | ❌ Not required | +| vLLM | Apache 2.0 | ✅ Required | ✅ Yes | ✅ Allowed | ❌ Not required | + +### 2.4 Comparison of Major Open Source AI Model Licenses + +| Model | License | Commercial use | Derivative models | Main restrictions | +|-----|---------|:--------:|:--------:|-------------| +| Llama 3.1 (8B/70B) | Meta Llama Community License | ✅ Conditional | ✅ Allowed | Permission from Meta required above 700 million MAU | +| Llama 3.1 (405B) | Meta Llama Community License | ✅ Conditional | ✅ Allowed | Restrictions on redistributing model weights | +| Mistral 7B | Apache 2.0 | ✅ Allowed | ✅ Allowed | No restrictions | +| Falcon 180B | Falcon License | ✅ Conditional | ✅ Allowed | License required when generating revenue | +| Gemma 2 | Gemma ToU | ✅ Conditional | ✅ Allowed | Compliance with Google's usage policy | +| Phi-3 | MIT | ✅ Allowed | ✅ Allowed | No restrictions | +| BERT, GPT-2 | Apache 2.0 / MIT | ✅ Allowed | ✅ Allowed | No restrictions | + +{{% alert title="Caution: recheck model licenses regularly" color="warning" %}} +The terms of AI model licenses can change when versions are updated. +If you use a model for a long period, **recheck each quarter whether the license terms have changed**. +Models with commercial use restrictions or MAU conditions in particular need to be reviewed again as the scale of the service changes. +{{% /alert %}} + +**Checkpoints: §8.5** +- [ ] Have the licenses of all frameworks used in developing the AI system been reviewed? +- [ ] Have the license terms (commercial use, MAU limits, and so on) of the pre-trained models in use been confirmed? +- [ ] Are license obligations (copyright notices and so on) included in the distributed artifacts? +- [ ] Has an SCA scan been performed to identify security vulnerabilities? + +--- + +## 3. §8.6 AI Data Management: Training Dataset Licenses ★ + +If an open data license applies to a dataset used to train an AI system, +the terms of that license must be fulfilled. + +### 3.1 Obligations of Major Open Data Licenses + +| License | Attribution | ShareAlike | Commercial use | How to comply | +|---------|:---------:|:--------:|:--------:|---------| +| CC0 1.0 | ❌ Not required | ❌ Not required | ✅ Allowed | No action needed | +| CC-BY 4.0 | ✅ Required | ❌ Not required | ✅ Allowed | State the data source in the model card | +| CC-BY-SA 4.0 | ✅ Required | ✅ Required | ✅ Allowed | Legal review required (license of the derivative model) | +| CC-BY-NC 4.0 | ✅ Required | ❌ Not required | ❌ Not allowed | Use only for non-commercial research purposes | +| ODbL (OpenStreetMap and others) | ✅ Required | ✅ Required | ✅ Allowed | Publish derivative databases under the same terms | + +### 3.2 Writing a Model Card + +If a CC-BY family dataset was used for training, state the source in the model card. + +```markdown +## Model Card: [Model name] + +### Training Data + +This model was trained using the following datasets: + +| Dataset | License | Source | +|---------|---------|------| +| Wikipedia (2024-01) | CC-BY-SA 4.0 | https://dumps.wikimedia.org/ | +| Common Crawl (CC-MAIN-2023) | No conditions | https://commoncrawl.org/ | +| [Internal dataset] | Owned by the organization | Internal | +``` + +**Checkpoints: §8.6** +- [ ] Are the list of training datasets and their licenses recorded in the AI SBOM? +- [ ] When CC-BY family data is used, is the source stated in the model card or public documentation? +- [ ] Is CC-BY-NC data not being used in commercial services? +- [ ] If CC-BY-SA data was used for training, has the license of the derivative model been discussed with the legal team? + +--- + +## 4. References + +- [Operation section home](../) +- [AI SBOM Guide](../2-ai-sbom/) +- [AI Supply Chain Verification](../3-supply-chain/) +- [ISO/IEC 5230: §3.3 Content Review](../../../iso5230_guide/3-content-review/) +- [ISO/IEC 18974: §4.3.2 Security Assurance](../../../iso18974_guide/3-content-review/2-security-assurance/) +- [Enterprise Open Source Management Guide: AI Compliance](../../../opensource_for_enterprise/7-ai-compliance/) diff --git a/content/en/guide/iso42001_guide/4-operation/2-ai-sbom/_index.md b/content/en/guide/iso42001_guide/4-operation/2-ai-sbom/_index.md new file mode 100644 index 0000000000..29d59b08d0 --- /dev/null +++ b/content/en/guide/iso42001_guide/4-operation/2-ai-sbom/_index.md @@ -0,0 +1,389 @@ +--- +title: "AI SBOM" +weight: 20 +type: docs +categories: ["guide"] +tags: ["ISO/IEC 42001", "AI SBOM", "SPDX 3.0", "AI Documentation"] +description: > + Describes how to build an AI SBOM (AI System Bill of Materials) and which tools generate it, + in accordance with ISO/IEC 42001 §7.5 (Documented information) and §8.5 (AI system life cycle). +--- + +## 1. What is an AI SBOM? + +An **AI SBOM (AI System Bill of Materials)** documents a list of every element that makes up an AI system, +along with its source and license. It extends the concept of the software SBOM (ISO/IEC 5230 §3.3.1) +to AI systems and is the key deliverable for the ISO/IEC 42001 §7.5 (Documented information) requirement. + +### Why an AI SBOM Is Needed: Transparency and Regulatory Compliance + +The AI SBOM is a key means of securing **transparency and explainability of AI systems**. +ISO/IEC 42001 Appendix C.2.11 states transparency and explainability as key objectives of the AI management system. + +| Requiring party | Basis | Role of the AI SBOM | +|---------|------|------------| +| **ISO/IEC 42001** | Appendix C.2.11 (transparency and explainability) | Securing transparency of AI system composition | +| **EU AI Act** | Technical documentation requirements for high-risk AI systems | Stating training data and model sources | +| **EU Cyber Resilience Act (CRA)** | Transparency obligations for AI products | Component list and vulnerability tracking | +| **OpenChain AI Work Group** | AI SBOM Compliance Guide | Standard procedures for AI SBOMs in the supply chain | + +{{% alert title="ISO/IEC 42003 and the AI SBOM" color="info" %}} +**ISO/IEC 42003** (in development), the implementation guidance for ISO/IEC 42001, is expected to include specific requirements related to AI SBOMs. +The OpenChain AI Work Group is working to have its AI SBOM guide reflected in +§6.2 (AI objectives) and Appendix C.2.11 (transparency and explainability) of 42003. +{{% /alert %}} + +### AI SBOM vs Software SBOM + +| Item | Software SBOM | AI SBOM | +|---------|---------------|---------| +| **Scope** | Libraries, packages, executables | Frameworks + models + datasets | +| **License types** | SPDX standard licenses | SPDX + custom AI licenses (Llama, Gemma ToU, and others) | +| **Standard format** | SPDX 2.x, CycloneDX 1.4 | SPDX 3.0 AI profile, CycloneDX 1.6 | +| **Additional information** | None | Number of model parameters, training data sources, model purpose, and so on | +| **Related ISO standard** | ISO/IEC 5230 §3.3.1 | ISO/IEC 42001 §7.5 | + +--- + +## 2. Components of an AI SBOM + +``` +AI SBOM + │ + ├── 1. AI frameworks and libraries + │ (same items as a regular software SBOM) + │ - name, version, license, PURL, hash + │ + ├── 2. Pre-trained Model + │ - name, version, license + │ - primaryPurpose (inference / training / fine-tuning / evaluation) + │ - modelCard URL (link to the model card) + │ - baseModel (original model, for fine-tuning) + │ + ├── 3. Training datasets + │ - name, version, license + │ - dataCollectionProcess (collection method) + │ - dataType (text / image / audio / etc.) + │ - knownBias (known bias, if any) + │ + ├── 4. Fine-tuning data (if applicable) + │ - name, license, source + │ - dataType + │ + └── 5. Fact Sheet (AI system card) ★ + - Bias information and limitations + - Data sources and data availability + - Summary of model performance and evaluation results + - Intended uses and prohibited uses + - Known safety and security issues +``` + +{{% alert title="What is a Fact Sheet?" color="info" %}} +A **Fact Sheet** (or extended model card) is a document that provides transparency and explainability for an AI system. +It is a key means of meeting ISO/IEC 42001 Appendix C.2.11 and the technical documentation requirements for high-risk AI under the EU AI Act. +The OpenChain AI Work Group is working to include this Fact Sheet information as a standard component of the AI SBOM. +{{% /alert %}} + +--- + +## 3. AI SBOM Standard Format Specifications (SPDX 3.0, CycloneDX 1.6) + +Two AI SBOM standard formats have become de facto industry standards. The two formats complement each other, +and an organization can adopt either one or both. + +| Standard | Released | AI specification | Key strengths | +|------|------|------------|----------| +| **SPDX 3.0** | 2024-04 | AI Profile (`AIPackage` class) | Strong expression of licenses and copyright, successor to ISO/IEC 5962:2021 | +| **CycloneDX 1.6** | 2024-04 | ML-BOM (`modelCard`) | Rich security, ethics, and performance metadata, ECMA-424 standard | + +The AI Work Group of the OpenChain Korea Work Group has developed an AI SBOM guide that supports both formats. + +### 3.1 Twelve Core Fields of the SPDX 3.0 AI Profile + +The SPDX 3.0 AI Profile expresses AI model metadata through the [`AIPackage`](https://spdx.github.io/spdx-spec/v3.0.1/model/AI/Classes/AIPackage/) +class. Among the AI-specific fields, the twelve core fields used most often are as follows. + +| # | Field | Meaning | Example value | +|---|------|------|--------| +| 1 | `typeOfModel` | Model type | "LLM", "Image Classification", "Reinforcement Learning" | +| 2 | `modelArchitecture` | Details of the model architecture | "Transformer Decoder", "Vision Transformer", "MoE" | +| 3 | `hyperparameter` | Hyperparameters (key/value) | `contextWindow: 131072`, `learningRate: 2e-5` | +| 4 | `modelDataPreprocessing` | Training data preprocessing method | "BPE Tokenization", "Image normalization (224×224)" | +| 5 | `informationAboutTraining` | Information about the training procedure | "Pretrained on 15T tokens, fine-tuned on KoAlpaca" | +| 6 | `informationAboutApplication` | Information about the intended application | "Customer support chatbot, EN/KO" | +| 7 | `metric` | Performance metrics | "MMLU: 0.78", "HumanEval: 0.65" | +| 8 | `metricDecisionThreshold` | Decision threshold | "Toxicity > 0.7 → reject" | +| 9 | `limitation` | Known limitations and weaknesses | "Hallucinates on dates after 2024-12" | +| 10 | `safetyRiskAssessment` | Result of the safety and security risk assessment | "Prompt injection assessment completed (PASS)" | +| 11 | `modelExplainability` | Explainability techniques | "SHAP analysis supported", "Attention map visualization" | +| 12 | `autonomyType` | Level of autonomy | "Human-in-the-loop", "Fully autonomous" | + +Supplementary fields (as needed): `domain` (application domain), `energyConsumption` (energy consumption), +`standardCompliance` (list of standards complied with), `useSensitivePersonalInformation` (whether sensitive personal information is used). +Inherited fields (`name`, `packageVersion`, `downloadLocation`, `suppliedBy`, `primaryPurpose`, +`releaseTime`) are written in the same way as in a regular SBOM. + +### 3.2 Example of an SPDX 3.0 AI SBOM + +```yaml +spdxVersion: SPDX-3.0 +SPDXID: SPDXRef-DOCUMENT +name: "MyAI-Service AI SBOM" +dataLicense: CC0-1.0 +created: "2026-04-01T00:00:00Z" +createdBy: + - type: Tool + identifier: "cdxgen-10.0" + +packages: + # 1. AI frameworks + - SPDXID: SPDXRef-pytorch + name: torch + versionInfo: "2.2.0" + supplier: "Organization: Meta AI" + downloadLocation: "https://pypi.org/project/torch/2.2.0/" + licenseConcluded: BSD-3-Clause + primaryPurpose: library + + - SPDXID: SPDXRef-transformers + name: transformers + versionInfo: "4.40.0" + supplier: "Organization: Hugging Face" + downloadLocation: "https://pypi.org/project/transformers/4.40.0/" + licenseConcluded: Apache-2.0 + primaryPurpose: library + + # 2. Pre-trained model + - SPDXID: SPDXRef-llama3-8b + name: meta-llama/Llama-3.1-8B-Instruct + versionInfo: "3.1" + supplier: "Organization: Meta AI" + downloadLocation: "https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct" + licenseConcluded: LicenseRef-Meta-Llama-Community-License + primaryPurpose: inference + # AI profile extension fields + modelCard: "https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct" + hyperparameter: + contextWindow: 131072 + + # 3. Fine-tuning dataset + - SPDXID: SPDXRef-dataset-alpaca-korean + name: beomi/KoAlpaca-v1.1a + versionInfo: "1.1a" + downloadLocation: "https://huggingface.co/datasets/beomi/KoAlpaca-v1.1a" + licenseConcluded: CC-BY-NC-4.0 + primaryPurpose: trainData + dataCollectionProcess: "Korean instruction-following dataset" + +relationships: + - spdxElementId: SPDXRef-llama3-8b + relationshipType: TRAINED_ON + relatedSpdxElement: SPDXRef-dataset-alpaca-korean +``` + +{{% alert title="Caution with CC-BY-NC datasets" color="warning" %}} +The `beomi/KoAlpaca-v1.1a` dataset in the example above is licensed under **CC-BY-NC 4.0** +and may be used only for **non-commercial purposes**. If a commercial service uses a model +fine-tuned with this dataset, consult the legal team immediately. + +This example is intended to explain the AI SBOM format. Commercial services should use datasets under licenses +that permit commercial use, such as Apache 2.0, MIT, CC-BY 4.0, or CC0. +{{% /alert %}} + +### 3.3 Core Fields of the CycloneDX 1.6 ML-BOM (Four Areas of modelCard) + +[CycloneDX 1.6 ML-BOM](https://cyclonedx.org/capabilities/mlbom/) expresses AI model metadata in the +`modelCard` field of a component. modelCard consists of four areas. + +**(1) `modelParameters`: Model configuration (7 fields)** + +| Field | Meaning | +|------|------| +| `approach` | Training approach (Supervised / Unsupervised / Reinforcement / Semi-supervised) | +| `task` | Type of model task (classification, regression, text-generation, and so on) | +| `architectureFamily` | Architecture family (transformer, CNN, RNN, LSTM, and so on) | +| `modelArchitecture` | Specific architecture (GPT-4o, ResNet-50, YOLOv8, and so on) | +| `datasets` | List of references to datasets used for training and evaluation | +| `inputs` | Input format (MIME type, tensor shape, and so on) | +| `outputs` | Output format | + +**(2) `quantitativeAnalysis`: Performance analysis (2 fields)** + +| Field | Meaning | +|------|------| +| `performanceMetrics` | Performance metrics (accuracy, F1, MSE, top-k, and so on) | +| `graphics` | Attached graphs and images of evaluation results | + +**(3) `considerations`: Ethical and operational considerations (7 fields)** + +| Field | Meaning | +|------|------| +| `users` | Intended users | +| `useCases` | Intended use cases | +| `technicalLimitations` | Technical limitations | +| `performanceTradeoffs` | Performance and accuracy trade-offs | +| `ethicalConsiderations` | Assessment of ethical risks | +| `environmentalConsiderations` ★ | Environmental impact (power consumption, carbon emissions), **new in 1.6** | +| `fairnessAssessments` | Fairness assessment (benefits and risks for each affected group) | + +**(4) `properties`: Free-form key-value** + +Additional internal organizational metadata (approver, review date, and so on) can be recorded. + +### 3.4 Example of a CycloneDX 1.6 ML-BOM + +```json +{ + "bomFormat": "CycloneDX", + "specVersion": "1.6", + "components": [ + { + "type": "machine-learning-model", + "bom-ref": "model-llama3.1-8b", + "name": "Meta Llama 3.1 8B Instruct", + "version": "3.1", + "supplier": { "name": "Meta AI" }, + "licenses": [ + { "license": { "name": "Llama 3.1 Community License" } } + ], + "modelCard": { + "bom-ref": "card-llama3.1-8b", + "modelParameters": { + "approach": { "type": "supervised" }, + "task": "text-generation", + "architectureFamily": "transformer", + "modelArchitecture": "Decoder-only Transformer", + "datasets": [ { "ref": "ds-meta-pretrain-15T" } ], + "inputs": [ { "format": "text/plain" } ], + "outputs": [ { "format": "text/plain" } ] + }, + "quantitativeAnalysis": { + "performanceMetrics": [ + { "type": "MMLU", "value": "0.78" }, + { "type": "HumanEval", "value": "0.65" } + ] + }, + "considerations": { + "users": [ "Enterprise developers" ], + "useCases": [ "Customer support chatbot (EN/KO)" ], + "technicalLimitations": [ "Hallucinates on dates after 2024-12" ], + "ethicalConsiderations": [ + { "name": "Prompt injection", "mitigation": "Input filtering" } + ], + "environmentalConsiderations": { + "energyConsumptions": [ + { + "activity": "inference", + "energyMeasure": { "value": 0.5, "unit": "kWh" } + } + ] + } + } + } + } + ] +} +``` + +{{% alert title="Recommendation to publish SPDX 3.0 and CycloneDX 1.6 together" color="success" %}} +To meet both the technical documentation requirements for high-risk systems under the EU AI Act and future ISO/IEC 42003 requirements, +it is recommended to **publish both** SPDX 3.0 (for license tracking) and the CycloneDX 1.6 ML-BOM (for security and ethics tracking). +The two formats can be linked through component IDs (PURL or `modelCard.bom-ref`). +{{% /alert %}} + +--- + +## 4. AI SBOM Generation Tools + +### 4.1 cdxgen (CycloneDX Format) + +cdxgen analyzes the dependencies of a Python AI project and generates an SBOM in CycloneDX format. +AI-specific fields must be added separately. + +```bash +# Generate an SBOM for a Python AI project +cdxgen -t python ./my-ai-project -o ai-sbom.json + +# Generate based on requirements.txt +cdxgen -t pypi -r requirements.txt -o ai-sbom.json +``` + +Tool guide: [cdxgen usage guide](../../../tools/5-cdxgen/) + +### 4.2 Syft (Container Image Analysis) + +If an AI service is deployed as a container, use Syft to generate an SBOM for the entire container image. + +```bash +# Generate an SBOM for an AI service container image +syft my-ai-service:latest -o spdx-json=ai-container-sbom.json +``` + +Tool guide: [Syft usage guide](../../../tools/6-syft/) + +### 4.3 Dependency-Track (AI SBOM Management and Vulnerability Tracking) + +Uploading the generated AI SBOM to Dependency-Track enables continuous vulnerability monitoring. + +```bash +# Upload an AI SBOM to Dependency-Track +curl -X "POST" "http://dtrack:8080/api/v1/bom" \ + -H "X-Api-Key: ${DTRACK_API_KEY}" \ + -F "bom=@ai-sbom.json" \ + -F "projectName=MyAI-Service" \ + -F "projectVersion=1.0.0" +``` + +Tool guide: [Dependency-Track usage guide](../../../tools/7-dependency-track/) + +--- + +## 5. AI SBOM Operating Procedure + +### Managing the AI SBOM Life Cycle + +``` +Start of AI system development + └─ Draft the AI SBOM (when frameworks and models are decided) + └─ Update immediately when components are added during development + └─ Final review of the AI SBOM before deployment + └─ Archive the AI SBOM at deployment (by version) + └─ Respond using the AI SBOM when new vulnerabilities are found during operation + └─ Monitor model license changes → update the AI SBOM as needed +``` + +**Checkpoints**: +- [ ] Have AI SBOMs been generated and archived for all deployed versions? +- [ ] Does the AI SBOM include all frameworks, models, and datasets? +- [ ] Is the license of each component recorded accurately? +- [ ] Is vulnerability monitoring performed based on the AI SBOM? +- [ ] Are model signing (OpenSSF Model Signing, Sigstore) verification results recorded in the AI SBOM? +- [ ] Is the training build provenance (SLSA for AI L1 or higher) kept as an attestation? + +{{% alert title="Model signing and build provenance" color="info" %}} +Sign model weight files with OpenSSF Model Signing (the `model-signing` CLI), +and attach provenance to training and fine-tuning builds according to the SLSA for AI level (L1 to L3). +For specific procedures, see [Supply Chain Verification §6.3 OpenSSF Model Signing](../3-supply-chain/#63-adopting-openssf-model-signing) and +[§6.4 SLSA for AI Build Levels](../3-supply-chain/#64-slsa-for-ai-build-levels). +{{% /alert %}} + +--- + +## 6. Connection with the AI Work Group Guide + +The AI Work Group of the OpenChain Korea Work Group has developed and published the **AI SBOM Compliance Guide**. +This guide explains in detail how to write AI SBOMs using the SPDX 3.0 AI profile. + +- AI Work Group activities and guide: [Resources: AI Work Group](../../../../resource/AI_work_group/) + +--- + +## 7. References + +- [Operation section home](../) +- [ISO/IEC 5230: §3.3.1 SBOM](../../../iso5230_guide/3-content-review/1-sbom/) +- [ISO/IEC 18974: §4.3.1 SBOM](../../../iso18974_guide/3-content-review/1-sbom/) +- [Tools: cdxgen](../../../tools/5-cdxgen/) +- [Tools: Syft](../../../tools/6-syft/) +- [Tools: Dependency-Track](../../../tools/7-dependency-track/) +- [AI Work Group](../../../../resource/AI_work_group/) diff --git a/content/en/guide/iso42001_guide/4-operation/3-supply-chain/_index.md b/content/en/guide/iso42001_guide/4-operation/3-supply-chain/_index.md new file mode 100644 index 0000000000..8e38d751db --- /dev/null +++ b/content/en/guide/iso42001_guide/4-operation/3-supply-chain/_index.md @@ -0,0 +1,357 @@ +--- +title: "AI Supply Chain Verification" +weight: 30 +type: docs +categories: ["guide"] +tags: ["ISO/IEC 42001", "AI Supply Chain", "External AI Models", "Open Source Verification"] +description: > + Describes how to verify the supply chain of open source AI models and AI services procured externally, + in accordance with ISO/IEC 42001 §8.8 (Use of AI systems supplied by external parties). +--- + +## 1. Overview + +ISO/IEC 42001 §8.8 requires appropriate evaluation and verification when using AI systems (models, APIs, services) +procured externally. From an open source perspective, the key is the procedure for verifying license, security, and supply chain risks +when procuring **external open source AI models**. + +--- + +## 2. Three Types of External AI Procurement + +| Type | Examples | Open source relevance | +|------|------|--------------| +| **Direct use of open source AI models** | Downloading Llama, Mistral, or Falcon model weights | High: the license applies directly | +| **AI services based on open source** | Hugging Face Inference API, Ollama | Medium: the license of the underlying model must be checked | +| **Commercial AI APIs** | OpenAI API, Google Vertex AI | Low: terms of service apply (OSS licenses do not apply directly) | + +Types 1 and 2 are areas where open source licenses apply directly and are the main subjects of §3 to §4. +Type 3 (commercial AI APIs) is also required in the same way by ISO/IEC 42001 §8.8, so it is covered separately in §5. + +--- + +## 3. Verification Checklist Before Procuring Open Source AI Models + +Verify the following items before adopting an external open source AI model. + +### 3.1 License Verification + +```markdown +## Open Source AI Model License Verification Checklist + +### Basic license information +- [ ] License type confirmed: ___________________ + (Apache 2.0 / MIT / Llama Community / Gemma ToU / other) +- [ ] Source URL of the license text: ___________________ +- [ ] License version confirmed (may differ from earlier versions of the same model) + +### Commercial use conditions +- [ ] Whether commercial use is permitted: ✅ Permitted / ⚠️ Conditional / ❌ Not permitted +- [ ] Conditions limiting the number of users (MAU): ___________________ + (for example, Llama 3: permission from Meta required above 700 million MAU) +- [ ] Revenue-based restrictions: ___________________ + +### Derivative (fine-tuning) conditions +- [ ] Whether fine-tuning is permitted: ✅ Permitted / ⚠️ Conditional / ❌ Not permitted +- [ ] Whether there is an obligation to disclose fine-tuned models: ___________________ +- [ ] License requirements for fine-tuned models: ___________________ + +### Redistribution conditions +- [ ] Whether redistribution of model weights is permitted: ✅ Permitted / ⚠️ Conditional / ❌ Not permitted +- [ ] Obligation to include license documents when redistributing: ___________________ + +### Attribution obligations +- [ ] Whether attribution is required: ✅ Required / ❌ Not required +- [ ] Attribution method and location: ___________________ + (service UI, documentation, API responses, and so on) + +### Whether legal review is required +- [ ] Legal team review completed if not a standard SPDX license: ✅ / Not applicable +- [ ] Date of legal team review: ___________________ +- [ ] Review opinion: ___________________ +``` + +### 3.2 Security Verification + +```markdown +### Security verification items +- [ ] Download from an official distribution channel confirmed + (official GitHub, official Hugging Face account) +- [ ] Exact match with the official namespace confirmed (typo-squatting defense; see §6.1 for detailed cases and defensive controls) +- [ ] File hash (SHA256) verified and recorded in the AI SBOM +- [ ] Model weight format confirmed + (Safetensors preferred; `.pt` and `.bin` pickle formats are verified in an isolated environment before use in production) +- [ ] Known vulnerabilities (CVEs) looked up + (NVD, OSV.dev search results: ___________________) +- [ ] Model weights reviewed for inserted malicious code or backdoors + (do not use models from untrusted sources; check for backdoor triggers with evaluation datasets) +- [ ] Registered with a channel for monitoring license changes + (GitHub Watch, official newsletter, and so on) +``` + +### 3.3 Supply Chain Risk Assessment + +```markdown +### Supply chain risk assessment items +- [ ] Trustworthiness of the model supplier confirmed + (individual / research institution / company; reputation in the open source community) +- [ ] Maintenance activity of the model confirmed + (date of last update, status of issue responses) +- [ ] License change history confirmed + (whether license terms have changed in the past) +- [ ] Alternative models identified + (alternatives in case of a license change or service discontinuation) +``` + +--- + +## 4. Summary of License Risks of Major Open Source AI Models + +| Model family | License | Main risks | +|---------|---------|-----------| +| **Llama 3.x** | Meta Llama Community License | Meta approval required above 700 million MAU, differences in terms between license versions | +| **Llama 2** | Meta Llama 2 Community License | Restrictions on use by competitors (from Meta's standpoint), restrictions on using the "Llama 2" name for derivative models | +| **Gemma 2** | Google Gemma ToU | License terminates immediately upon violation of Google's usage policy | +| **Falcon-7B / 40B** | Apache-2.0 (since 2023-09) | Low risk: commercial use is free | +| **Falcon-180B** | TII Falcon LLM License v1.0 | Conditional commercial use, such as revenue thresholds; the original terms must be checked directly | +| **Mistral 7B** | Apache 2.0 | Low risk | +| **Phi-3** | MIT | Low risk | +| **GPT-2, BERT** | MIT / Apache 2.0 | Low risk | + +{{% alert title="Caution when using models under custom licenses" color="warning" %}} +For models under custom licenses such as Llama and Gemma, **read the license text directly** +and use the model only after a legal team review. Custom licenses are not SPDX standard licenses, +so they cannot be reviewed automatically by common license classification tools. +{{% /alert %}} + +--- + +## 5. §8.8 Evaluation Checklist for Commercial AI APIs + +Open source licenses do not apply directly to commercial AI APIs such as OpenAI, Anthropic, Google Vertex AI, AWS Bedrock, and Azure OpenAI, +but ISO/IEC 42001 §8.8 equally requires evaluation of externally supplied +AI systems. Review the following three areas before adoption. + +### 5.1 Verifying Data Processing and Use for Training + +```markdown +### Data processing verification items +- [ ] Whether input data is used for training (opt-in / opt-out policy) + (confirm whether Enterprise/API plans are opt-out by default) +- [ ] Retention period of output data (whether a zero-retention option is available) +- [ ] Location (country) where API call logs are stored: review cross-border transfer under the Personal Information Protection Act (개인정보보호법) +- [ ] Whether the data processing region can be selected (EU or KR regions, and so on) +- [ ] Whether a BAA (Business Associate Agreement) can be signed (healthcare sector) +- [ ] Whether a policy blocking input of sensitive information and trade secrets has been established +``` + +### 5.2 Comparison of IP Indemnification + +To address the risk of copyright infringement by generative AI output, major providers operate IP indemnification policies. +Coverage and conditions differ by plan, so check the **original terms** and the **latest policy pages** directly. + +| Provider | Name of indemnification policy | Conditions (summary) | +|--------|--------------|---------------| +| **OpenAI** | Copyright Shield | ChatGPT Enterprise, Team, and API users (ChatGPT Plus and free plans explicitly excluded) | +| **Anthropic** | IP Indemnification | When subscribed under the Commercial Service Agreement | +| **Google Cloud** | Generative AI Indemnification | Models named in the Vertex AI terms, when the terms are complied with | +| **AWS** | IP Indemnification | Bedrock Titan + third-party models named in the terms (some, such as Claude and Llama; check the original terms), Amazon Q | +| **Microsoft (Azure OpenAI)** | Customer Copyright Commitment | Azure OpenAI Service | +| **Microsoft (M365 Copilot)** | Copilot Copyright Commitment | M365 Copilot, GitHub Copilot Business/Enterprise | + +Common indemnification requirements: content filters enabled, post-review of output, no intentional attempts at infringement, +and lawful rights to input data (assurance that the copyright and license of materials the user enters in the prompt are lawful). +The scope of damages that can be claimed under indemnification (legal advisory costs, settlements, and so on) differs by provider, so consult the legal team in advance. +Actual terms change frequently, so check the **original terms** directly before adoption. + +### 5.3 Monitoring Changes to Terms of Service and Availability + +```markdown +### Terms and availability monitoring items +- [ ] Registered with a channel for notifications of terms changes (email notifications, RSS of the change log page) +- [ ] Price change policy confirmed (contractual advance notice period) +- [ ] SLA availability guarantee level (for example, 99.9%) +- [ ] End-of-service (EOL) policy: secure a migration period +- [ ] Alternative providers identified (preventing vendor lock-in; maintain a mapping table of equivalent models) +- [ ] Review of clauses limiting liability for output (hallucination and bias disclaimer clauses) +``` + +{{% alert title="Recommendation to operate commercial APIs and open source models in parallel" color="success" %}} +To prepare for availability, terms change, and pricing risks of commercial APIs, it is recommended to **secure a self-hosted open source model option +as a backup**. Recording the model mapping for both options together in the AI SBOM +enables a quick switchover. +{{% /alert %}} + +--- + +## 6. Defending Against Model Supply Chain Attacks + +Since 2024, open source model weight files and model hubs (Hugging Face, PyTorch Hub, and others) have been identified as a **new +attack surface**. Before adoption, be aware of the following attack types and apply defensive controls. + +### 6.1 Known Attack Types + +| Attack | Description | Defense | +|------|------|------| +| **Pickle RCE** | Arbitrary code inserted into models serialized with Python pickle, such as PyTorch `.pt` and `.bin`. Executed when the model is loaded | Prefer the Safetensors format, load untrusted models in an isolated container | +| **Typo-squatting** | Publishing misspelled variants of model or package names on Hugging Face or PyPI (for example, `meta-Ilama` vs `meta-llama`) | Explicit verification of the official namespace, pinning PURLs, hash verification at download | +| **Model Poisoning** | Backdoors inserted into training data or weights, producing malicious output when a specific trigger input is given | Trusted providers, comparison against benchmark datasets, checking for trigger patterns | +| **License-flip** | Changing the model license after the fact and attempting to apply it retroactively to existing users | Keep the license text and weight hash at the time of download together | +| **Dependency Confusion** | Publishing a public model with the same name as a private model to induce incorrect downloads | Explicitly specify the namespace, set private registry priority | + +### 6.2 Recommended Defensive Controls + +```markdown +### Model supply chain defensive controls checklist +- [ ] Prefer the Safetensors format for model weights (avoid pickle) +- [ ] Pin model file hashes (SHA-256) + record them in the AI SBOM +- [ ] Adopt OpenSSF Model Signing and Sigstore signature verification (§6.3) +- [ ] Assess the SLSA for AI build level (§6.4) +- [ ] Verify models first in an isolated environment (sandbox) before moving them to production +- [ ] Mandatory prior review by the security team when adopting new models +- [ ] Automate typo-squatting checks of model hub accounts (CI stage) +- [ ] Keep the model license text, weight hash, and download date together +``` + +{{% alert title="2024 Hugging Face Hub incident" color="warning" %}} +In February 2024, JFrog Security researchers identified **about 100 malicious models** on the Hugging Face Hub. +Many of them were designed to abuse pickle serialization to execute RCE payloads such as reverse shells when the model is loaded +(some were for system reconnaissance). Hugging Face had operated PickleScan and the Safetensors +format before the incident, but the incident led to strengthened security. **The official recommendation is to use the Safetensors format +from trusted sources**. +{{% /alert %}} + +### 6.3 Adopting OpenSSF Model Signing + +[OpenSSF Model Signing](https://github.com/sigstore/model-transparency) is an ML model signing standard +that supports three methods: Sigstore (keyless OIDC signing), X.509 certificates, and public keys. +As of 2026, the official package name is `model-signing` (v1.1+). + +**(1) Installation** + +```bash +# Basic installation (uses Sigstore) +pip install model-signing + +# Including X.509 (PKCS #11 HSM) support +pip install model-signing[pkcs11] +``` + +**(2) Signing a model (Sigstore keyless method, recommended)** + +```bash +# Compute a digest over the entire model directory and sign it with Sigstore +# → Sigstore Fulcio issues a short-lived certificate for the OIDC identity and records it in the Rekor transparency log +# → Creates the model.sig file +model_signing sign sigstore /path/to/llama-3.1-8b +``` + +**(3) Verifying a model (confirming the signer's identity)** + +```bash +# Example using Google OIDC +model_signing verify sigstore /path/to/llama-3.1-8b \ + --signature model.sig \ + --identity "release-bot@example.com" \ + --identity_provider "https://accounts.google.com" + +# Example using GitHub Actions OIDC (most common in CI environments) +model_signing verify sigstore /path/to/llama-3.1-8b \ + --signature model.sig \ + --identity "https://github.com/myorg/myrepo/.github/workflows/release.yml@refs/heads/main" \ + --identity_provider "https://token.actions.githubusercontent.com" +``` + +> Recheck the exact CLI syntax in the official [`sigstore/model-transparency`](https://github.com/sigstore/model-transparency) README at the time of adoption (subcommand and option names may change between versions). + +**(4) Using your own keys (offline environments)** + +```bash +# Signing (your own key pair, such as RSA/EC) +model_signing sign key /path/to/model --private_key signing.key + +# Verification +model_signing verify key /path/to/model --signature model.sig --public_key signing.pub +``` + +**(5) Linking with the AI SBOM** + +Record the signature verification results (signer identity, issue date, Rekor log index) in the AI SBOM +`hashes` and `attestation` fields to secure audit traceability. In CycloneDX 1.6, they can be attached directly +to the `signature` field of the component. + +### 6.4 SLSA for AI Build Levels + +The build track of [SLSA (Supply-chain Levels for Software Artifacts)](https://slsa.dev/spec/v1.0/) +applies equally to ML model build pipelines. **SLSA for AI** is a use case that maps it +to the model training, fine-tuning, and conversion processes. + +| Level | Requirements (applied to ML models) | Use | +|------|----------------------|------| +| **L0** | (A convenience label for the state where SLSA is not applied; not a level defined by the standard) No provenance | Allowed only for local experiments and research | +| **L1** | The build platform automatically generates provenance (training data, code, hyperparameters, environment) | Internal models in general | +| **L2** | Provenance carries a tamper-resistant signature (in-toto attestation + Sigstore), using a hosted builder | Externally distributed models and models provided to customers | +| **L3** | Prevention of user-defined code injection (covert or explicit) during the build: blocking injection of training code and hyperparameters outside the build definition, hardened build environment | High-risk and regulated industries | + +The official stable version of SLSA is v1.0 (GA in 2023-04), and a v1.1 draft is in progress. At the time of adoption, +check the latest stable version in the [official SLSA specification](https://slsa.dev/spec/v1.0/). + +**Key items to include in ML model provenance** + +- Training data hash (dataset version, license) +- Training code git commit, CI/CD builder ID +- Hyperparameters, training environment (GPU type, library versions) +- Base model ID and hash (for fine-tuning) +- Build start and end times, builder identity + +**Recommended order of application** + +1. **Adopt L1 first**: Automatically generate provenance in the training pipeline (for example, a GitHub Actions builder automatically outputs an in-toto attestation) +2. **L2 stage**: Add a Sigstore signature to the provenance (using the tools in §6.3) +3. **L3 stage**: Adopt a hardened build environment (for example, isolated workflows on GitHub Actions hosted runners, isolated GCP/AWS builders) +4. Record the SLSA level in the AI SBOM (`properties.slsaLevel`) + +{{% alert title="Linking OpenSSF, Sigstore, and SLSA" color="success" %}} +OpenSSF Model Signing (signing), Sigstore (transparency log), and SLSA for AI (provenance) are +three complementary standards. Model distributors generate SLSA L2 or higher provenance as an in-toto attestation +and sign and register it with Sigstore using the model-signing CLI, and verifiers can then check for tampering +through the Rekor log. This combination is the de facto industry standard for ML supply chain security in 2026. +{{% /alert %}} + +--- + +## 7. External AI Model Procurement Process + +```mermaid +flowchart TD + A[Identify the model needed] --> B[List candidate models] + B --> C[License verification] + C --> D{Custom license?} + D -- Yes --> E[Legal team review] + D -- No --> F[Security verification] + E --> F + F --> G[Supply chain risk assessment] + G --> H[Approval for adoption] + H --> I[Register in AI SBOM] + I --> J[Register for license monitoring] +``` + +**Checkpoints**: +- [ ] Was the license verification procedure performed before adopting the external open source AI model? +- [ ] For models under custom licenses, has the legal team review been completed? +- [ ] Are the adopted external models registered in the AI SBOM? +- [ ] Is there a channel for monitoring license changes of external models? +- [ ] Have alternative models been identified for use in case license terms are violated? +- [ ] When adopting a commercial AI API, was the §5 evaluation (data processing, IP indemnification, terms monitoring) performed? +- [ ] Have the hash, format (Safetensors preferred), and namespace of the model weight files been verified? + +--- + +## 8. References + +- [Operation section home](../) +- [Open Source Management in AI Systems](../1-oss-in-ai/) +- [AI SBOM Guide](../2-ai-sbom/) +- [ISO/IEC 18974: §4.3.2 Security Assurance](../../../iso18974_guide/3-content-review/2-security-assurance/) +- [Enterprise Open Source Management Guide: AI Compliance](../../../opensource_for_enterprise/7-ai-compliance/) +- [AI Work Group](../../../../resource/AI_work_group/) diff --git a/content/en/guide/iso42001_guide/4-operation/_index.md b/content/en/guide/iso42001_guide/4-operation/_index.md new file mode 100644 index 0000000000..d1681af3e7 --- /dev/null +++ b/content/en/guide/iso42001_guide/4-operation/_index.md @@ -0,0 +1,66 @@ +--- +title: "4. Operation" +weight: 40 +type: docs +categories: ["guide"] +tags: ["ISO/IEC 42001", "AI Operation", "AI Life Cycle", "Open Source"] +description: > + Gives an overview of the requirements of ISO/IEC 42001 §8 (Operation) that intersect with open source management. + It points to detailed pages on open source compliance in the AI system life cycle, AI SBOMs, and AI supply chain verification. +--- + +## 1. Overview + +ISO/IEC 42001 §8 covers the **actual operation** stage of the AI management system. +From an open source perspective, §8 is the section with the most points of intersection. + +| Subclause | Description | Open source intersection | +|---------|------|-------------| +| §8.1 | Operational planning and control | Operating the open source review process | +| §8.2 to 8.3 | AI risk assessment and treatment (during operation) | Responding to OSS vulnerability risks | +| §8.4 | AI system impact assessment (during operation) | System impact of OSS vulnerabilities | +| **§8.5** | AI system life cycle | ★ OSS framework and model licenses | +| **§8.6** | Data for AI systems | ★ Open dataset licenses | +| §8.7 | Feedback interface | - | +| **§8.8** | Procurement of external AI systems | ★ Supply chain verification of external OSS models | + +--- + +## 2. Detailed Pages + +The open source intersection items of this section are covered in detail on the following pages: + +| Page | Target clauses | Key content | +|--------|---------|---------| +| [Open Source Management in AI Systems](./1-oss-in-ai/) | §8.5, §8.6 | License compliance for AI frameworks, models, and datasets | +| [AI SBOM](./2-ai-sbom/) | §7.5, §8.5 | Building AI SBOMs, generation tools, using SPDX 3.0 | +| [AI Supply Chain Verification](./3-supply-chain/) | §8.8 | Verification checklist for procuring external open source AI models | + +--- + +## 3. §8.1 Operational Planning: Integrating the Open Source Review Process + +Integrate open source compliance review steps into the AI system development process. + +```mermaid +flowchart TD + A[Planning stage] --> A1["List OSS frameworks and models\nPrior license review"] + A1 --> B[Development stage] + B --> B1["Check license when adding components\nDraft AI SBOM\nSCA scan"] + B1 --> C[Test and verification stage] + C --> C1["Review AI SBOM\nConfirm license obligations are fulfilled\nConfirm Critical CVEs are resolved"] + C1 --> D[Deployment stage] + D --> D1["Archive final AI SBOM\nConfirm license notices are included"] +``` + +**Checkpoints**: +- [ ] Does the AI system development process include open source compliance review steps? +- [ ] Is there a gate that prevents an AI system from being deployed without open source review? + +--- + +## 4. References + +- [ISO/IEC 42001 Guide home](../) +- [Enterprise Open Source Management Guide: 3. Process](../../opensource_for_enterprise/3-process/) +- [Enterprise Open Source Management Guide: AI Compliance](../../opensource_for_enterprise/7-ai-compliance/) diff --git a/content/en/guide/iso42001_guide/5-evaluation/_index.md b/content/en/guide/iso42001_guide/5-evaluation/_index.md new file mode 100644 index 0000000000..ea9bd6e5a3 --- /dev/null +++ b/content/en/guide/iso42001_guide/5-evaluation/_index.md @@ -0,0 +1,123 @@ +--- +title: "5. Performance Evaluation and Improvement" +weight: 50 +type: docs +categories: ["guide"] +tags: ["ISO/IEC 42001", "Performance Evaluation", "Improvement", "Monitoring"] +description: > + Explains the requirements of ISO/IEC 42001 §9 (Performance evaluation) and §10 (Improvement) that intersect with open source management. + It describes how to integrate open source compliance metrics with the performance indicators of the AI management system. +--- + +## 1. Overview + +ISO/IEC 42001 §9 and §10 cover the cycle of confirming whether the AI management system is achieving its intended results +and improving where it falls short. From an open source perspective, +§9.1 (Monitoring and measurement) is linked with open source compliance metrics. + +--- + +## 2. §9.1 Monitoring and Measurement: Open Source Compliance Metrics ★ + +ISO/IEC 42001 requires monitoring and measuring the performance of the AI management system. +Integrate existing open source compliance metrics into the performance indicators of the AI management system. + +### Examples of Open Source-Related AI Performance Indicators + +| Indicator | Measurement method | Target | Measurement frequency | +|------|---------|------|---------| +| **AI SBOM currency rate** | Rate of AI SBOM updates completed relative to AI deployments | 100% | At every deployment | +| **Model license review rate** | Rate of license reviews completed relative to newly adopted models | 100% | Quarterly | +| **Critical CVE patch rate** | Rate of Critical CVEs patched within 14 days of discovery | 95% or more | Monthly | +| **Dataset license recording rate** | Rate of dataset licenses recorded in the AI SBOM | 100% | Quarterly | +| **Number of OSS license violations** | Number of license violations found and time taken to resolve them | Target of 0 | Monthly | +| **AI supply chain verification rate** | Rate of supply chain verification completed when procuring external models | 100% | At every procurement | + +{{% alert title="Integrating with ISO/IEC 5230 and 18974 metrics" color="success" %}} +If you already operate open source compliance metrics based on ISO/IEC 5230 or 18974, +manage them in an integrated way by **adding AI-related metrics to the existing open source dashboard**. +There is no need to build a separate AI-only measurement framework. +{{% /alert %}} + +**Checkpoints**: +- [ ] Are performance indicators related to open source compliance of AI systems defined? +- [ ] Are the indicators measured and recorded at the defined frequency? +- [ ] Are improvement actions established when targets are not met? + +--- + +## 3. §9.2 Internal Audit + +Conduct regular internal audits to check whether the AI management system meets the ISO 42001 requirements. +Include open source compliance items in the internal audit checklist. + +### AI Open Source Internal Audit Checklist + +```markdown +## Internal Audit Checklist for Open Source Compliance of AI Systems +### Audit target: [AI system name] / Audit date: YYYY-MM-DD + +#### AI SBOM +- [ ] Does the AI SBOM reflect the latest deployed version? +- [ ] Are all frameworks, models, and datasets included in the AI SBOM? +- [ ] Is the license information of each component recorded accurately? + +#### License compliance +- [ ] Have the license terms of all open source AI components been reviewed? +- [ ] Have the license obligations (copyright notices, change notices, and so on) been fulfilled? +- [ ] Are models with commercial use restrictions not being used in commercial services? + +#### Security vulnerability management +- [ ] Has an SCA scan been performed within the last quarter? +- [ ] Have identified Critical/High CVEs been patched in a timely manner? +- [ ] Is the vulnerability handling history recorded? + +#### AI supply chain +- [ ] Has license verification of externally procured AI models been completed? +- [ ] Are license changes of external models being monitored? +``` + +--- + +## 4. §9.3 Management Review + +When top management reviews the AI management system, include the status of open source compliance in the reported items. + +**Examples of management review report items**: +- Status of AI SBOM management (currency rate, coverage) +- Occurrences of open source license violations and actions taken +- Status of AI system security vulnerabilities and patch rate +- Status of open source risks in the AI supply chain + +--- + +## 5. §10.1 Nonconformity and Corrective Action + +When an open source compliance violation is found, follow this procedure: + +1. **Immediate action**: Isolate the violation or take a temporary measure (for example, suspend the affected feature) +2. **Cause analysis**: Identify the root cause of why the violation occurred +3. **Corrective action**: Resolve the violation (license compliance, replacing the component, and so on) +4. **Preventing recurrence**: Improve the process (strengthening review procedures, training, and so on) +5. **Record keeping**: Document the violation, the actions, and the results + +--- + +## 6. §10.2 Continual Improvement + +Directions for continually improving the open source management framework for AI systems: + +- Update review criteria for new types of AI model licenses +- Adopt tools that automate AI SBOM generation (using cdxgen, Syft, and others) +- Reflect the latest guides from the AI Work Group + +--- + +## 7. References + +- [ISO/IEC 42001 Guide home](../) +- [Enterprise Open Source Management Guide: 6. Declaring Conformance](../../opensource_for_enterprise/6-conforming/) +- [ISO/IEC 5230: §3.6 Conformance](../../iso5230_guide/6-conformance/) +- [Tools: cdxgen (AI SBOM generation)](../../tools/5-cdxgen/) +- [Tools: Syft (container image analysis)](../../tools/6-syft/) +- [Tools: Dependency-Track (vulnerability tracking)](../../tools/7-dependency-track/) diff --git a/content/en/guide/iso42001_guide/_index.md b/content/en/guide/iso42001_guide/_index.md new file mode 100644 index 0000000000..0273ddb28b --- /dev/null +++ b/content/en/guide/iso42001_guide/_index.md @@ -0,0 +1,137 @@ +--- +title: "ISO/IEC 42001 Guide" +linkTitle: "ISO/IEC 42001 Guide" +weight: 30 +type: docs +categories: ["guide"] +tags: ["ISO/IEC 42001", "AI Management System", "AI Compliance", "AIMS"] +description: > + A guide that explains the ISO/IEC 42001 AI management system standard from an open source perspective. + It covers the key requirements where AI system development and operation intersect with open source compliance. +--- + +This guide explains, **from the perspective of open source practitioners**, the key requirements of ISO/IEC 42001 (AI Management System) +that intersect with open source management. It provides practical guidance on what must be complied with when using open source frameworks, +pre-trained models, and training datasets in AI systems, and on how to structure an AI SBOM. + +**Author : OpenChain Korea Work Group / [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)** + +## What is ISO/IEC 42001? + +ISO/IEC 42001:2023 is an international standard for an **AI management system (AIMS, AI Management System)** that enables organizations +to develop, operate, and manage AI systems responsibly and transparently. + +| Item | Description | +|------|------| +| **Official title** | ISO/IEC 42001:2023 Information technology - Artificial intelligence - Management system | +| **Developed by** | ISO/IEC JTC 1/SC 42 (Artificial intelligence) | +| **Year published** | 2023 | +| **Standard family** | ISO management system standards (same structure as ISO 9001 and ISO 27001) | +| **Applies to** | All organizations that develop, provide, or use AI products and services | + +{{% alert title="Differences from ISO/IEC 5230 and 18974" color="warning" %}} + +ISO/IEC 5230 (license compliance) and ISO/IEC 18974 (security assurance) were first developed as specifications by the **OpenChain Project** +and then adopted by ISO as standards. They are specialized for open source software management. +OpenChain provides free self-certification checklists for them. + +**ISO/IEC 42001** is an **AI management system** standard developed independently by ISO/IEC JTC 1 SC 42, +and it is unrelated to OpenChain. It is not specialized for open source but covers AI governance as a whole, +and no official self-certification checklist is provided for it. + +For a detailed comparison of the three standards, see the **[Standards Comparison](./compare/)** page. + +{{% /alert %}} + +## Positioning of This Guide + +{{% alert title="An ISO 42001 guide from an open source perspective" color="success" %}} + +ISO/IEC 42001 is a broad standard that covers the AI management system as a whole (risk, governance, life cycle, human oversight, and so on). +This guide focuses on the **requirements that intersect with open source management**. + +| What this guide covers | What this guide does not cover | +|----------------------|--------------------------| +| Open source license compliance for AI systems | AI ethics governance in general | +| Open source management of AI frameworks, models, and datasets | AI bias and fairness evaluation | +| Building and maintaining AI SBOMs | Human oversight and explainability requirements | +| Supply chain verification of open source AI models | AI system performance optimization | + +For AI compliance in general, see [Enterprise Open Source Management Guide: AI Compliance](../opensource_for_enterprise/7-ai-compliance/). + +{{% /alert %}} + +## How to Confirm Conformance with ISO/IEC 42001 + +Unlike OpenChain, ISO 42001 does not provide an official self-certification checklist. +An organization can confirm or declare conformance in the following three ways. + +{{% pageinfo %}} + +**Method 1. Self gap analysis (Self-assessment)** + +The organization internally reviews each clause of ISO 42001 (the "shall" requirements), evaluates its current level, +and establishes an improvement plan. It costs nothing, and the checkpoints in this guide can be used. + +--- + +**Method 2. Second-party assessment** + +A customer or partner company directly evaluates the organization's AI management system. +It is used to build trust in the supply chain. + +--- + +**Method 3. Third-party certification** + +An ISO certification body such as BSI or TÜV SÜD conducts an audit and issues an official ISO 42001 certificate. +It operates in the same way as ISO 27001 certification. + +{{% /pageinfo %}} + +## Structure of the Guide + +This guide follows the §4 to §10 structure of ISO/IEC 42001, focusing on the sections that intersect with open source. + +| Section | ISO 42001 clause | Open source intersection | +|------|----------------|----------------| +| [Standards Comparison](./compare/) | - | Relationship among ISO 5230, 18974, and 42001 | +| [1. Context of the Organization and Leadership](./1-context-leadership/) | §4, §5 | Including OSS principles in the AI policy | +| [2. Planning](./2-planning/) | §6 | AI risk assessment, OSS license and vulnerability risks | +| [3. Support](./3-support/) | §7 | AI OSS competence, AI SBOM documentation | +| [4. Operation](./4-operation/) | §8 | OSS frameworks, datasets, and supply chain | +| └ [Open Source Management in AI Systems](./4-operation/1-oss-in-ai/) | §8.5, §8.6 | OSS license compliance | +| └ [AI SBOM](./4-operation/2-ai-sbom/) | §7.5, §8.5 | Building and maintaining AI SBOMs | +| └ [AI Supply Chain Verification](./4-operation/3-supply-chain/) | §8.8 | Verifying procurement of external OSS models | +| [5. Performance Evaluation and Improvement](./5-evaluation/) | §9, §10 | OSS compliance metrics | + +## Connection with the AI Work Group + +The **AI Work Group** of the OpenChain Korea Work Group is developing an AI SBOM compliance guide. +The [AI SBOM](./4-operation/2-ai-sbom/) section of this guide is linked with the deliverables of the AI Work Group. + +- AI Work Group activities: [Resources: AI Work Group](../../resource/AI_work_group/) + +## Abbreviations + +| Abbreviation | Full name | Meaning | +|------|------|------| +| **AIMS** | AI Management System | AI management system (the core concept of ISO/IEC 42001) | +| **AI SBOM** | AI System Bill of Materials | List of AI system components (models, datasets, frameworks) | +| **ML-BOM** | Machine Learning Bill of Materials | ML model SBOM format of CycloneDX 1.6 | +| **AI Profile** | SPDX 3.0 AI Profile | AI SBOM specification based on the `AIPackage` class of SPDX 3.0 | +| **GPAI** | General-Purpose AI | General-purpose AI model category under EU AI Act §53 | +| **OSAID** | Open Source AI Definition | Open Source AI Definition by OSI (2024-10) | +| **MAU** | Monthly Active Users | Number of monthly active users (for example, the Llama license threshold) | +| **CRA** | EU Cyber Resilience Act | EU Cyber Resilience Act (entered into force 2024-12) | +| **NIS 2** | Network and Information Security Directive 2 | EU Network and Information Security Directive (applied from 2024-10) | +| **EO 14028** | US Executive Order 14028 | US Executive Order on Improving the Nation's Cybersecurity (2021-05) | +| **DPO** | Data Protection Officer | Data protection officer (Personal Information Protection Act (개인정보보호법)) | +| **EPSS** | Exploit Prediction Scoring System | Exploit prediction score (FIRST.org) | +| **KEV** | Known Exploited Vulnerabilities | Catalog of vulnerabilities exploited in the wild (CISA) | + +**Consistent license naming** (throughout the guide): +- Llama 3.x is written as "Meta Llama 3.x Community License" +- Gemma is written as "Gemma Terms of Use" +- Falcon is written as "Falcon License" (TII Falcon LLM License) +- Standard SPDX licenses use SPDX IDs (for example, `Apache-2.0`, `MIT`, `BSD-3-Clause`) diff --git a/content/en/guide/iso42001_guide/compare/_index.md b/content/en/guide/iso42001_guide/compare/_index.md new file mode 100644 index 0000000000..15f0e62121 --- /dev/null +++ b/content/en/guide/iso42001_guide/compare/_index.md @@ -0,0 +1,218 @@ +--- +title: "ISO 5230, 18974, and 42001 Comparison" +weight: 5 +type: docs +categories: ["guide"] +tags: ["ISO/IEC 5230", "ISO/IEC 18974", "ISO/IEC 42001", "Comparison"] +description: > + Compares the purpose, structure, and open source relevance of three standards: ISO/IEC 5230 (license compliance), + ISO/IEC 18974 (security assurance), and ISO/IEC 42001 (AI management system). +--- + +## 1. Relationship Among the Three Standards + +A company that develops AI systems using open source stands at the point where three standards intersect. + +```mermaid +flowchart LR + subgraph oss["Open source management"] + A["ISO/IEC 5230\nLicense compliance\n(licenses, SBOM, artifacts)"] + B["ISO/IEC 18974\nSecurity assurance\n(vulnerability detection, response, CVD)"] + end + + subgraph ai["AI system management"] + C["ISO/IEC 42001\nAI management system\n(governance, risk, life cycle)"] + end + + oss -- "Open source compliance\nwithin AI systems" --> ai +``` + +ISO 5230 and 18974 are standards for managing **open source itself**, +while ISO 42001 is a standard for managing **AI systems**. The two areas are independent, +but they intersect when AI systems use open source. + +{{% alert title="ISO/IEC 42003, AI system implementation guidance (in development)" color="info" %}} +While ISO/IEC 42001 sets out requirements as principles, +**ISO/IEC 42003** is guidance (a newly approved work item) that explains specific ways to implement 42001. +The OpenChain AI Work Group is working to have the **AI SBOM** reflected in the transparency and explainability clause of ISO 42003 +(Appendix C.2.11, Transparency and Explainability). +For the latest developments on mandatory AI SBOMs, see [AI Work Group activities](../../../resource/AI_work_group/). +{{% /alert %}} + +--- + +## 2. Comparison of Basic Information + +| Item | ISO/IEC 5230 | ISO/IEC 18974 | ISO/IEC 42001 | ISO/IEC 42003 | +|-----------|-------------|--------------|--------------|--------------| +| **Status** | Published (2020) | Published (2023) | Published (2023) | In development | +| **Developed by** | OpenChain Project → ISO | OpenChain Project → ISO | ISO JTC 1/SC 42 | ISO JTC 1/SC 42 | +| **Maintained by** | Linux Foundation OpenChain | Linux Foundation OpenChain | ISO | ISO | +| **Standard family** | OpenChain-specific standard | OpenChain-specific standard | ISO management system standard | ISO 42001 implementation guidance | +| **Scope** | Open source in the software supply chain | Open source security vulnerabilities | AI systems as a whole | How to implement ISO 42001 | +| **Main subject of management** | Fulfilling license obligations | CVE detection and response | AI governance and life cycle | Including AI SBOM and transparency | + +--- + +## 3. SC 42 Family Mapping + +ISO/IEC 42001 is not a standalone standard but the core of the family of AI standards developed by the +**ISO/IEC JTC 1/SC 42 (Artificial intelligence)** committee. Referring to the following standards together during certification audits +and practical application helps meet the requirements more fully. + +| Standard number | Title | Published | Role (relationship with 42001) | +|----------|------|------|---------------------| +| **ISO/IEC 22989:2022** | AI concepts and terminology | 2022 | Terminology baseline for 42001 | +| **ISO/IEC 23053:2022** | Framework for AI systems using ML | 2022 | Reference model for ML system structure | +| **ISO/IEC 23894:2023** | Guidance on risk management | 2023 | Specific guidance for 42001 §6.1 risk management | +| **ISO/IEC 38507:2022** | Governance implications of AI | 2022 | AI governance guidance for boards and executives | +| **ISO/IEC 5338:2023** | AI System life cycle processes | 2023 | Specific processes for the 42001 §8.5 AI life cycle | +| **ISO/IEC 5339:2024** | Guidance for AI applications | 2024 | Guidance for AI application developers | +| **ISO/IEC TR 5469:2024** | Functional safety and AI systems | 2024 | Safety-critical AI systems | +| **ISO/IEC 42001:2023** | AI Management System (AIMS) | 2023 | **Certifiable core standard** | +| **ISO/IEC 42003** | AIMS implementation guidance | In development | 42001 implementation guidance (efforts to include AI SBOM) | +| **ISO/IEC 42005:2025** | AI System Impact Assessment | 2025 | Specific method for the 42001 §6.1.4 impact assessment | +| **ISO/IEC 42006** | Bodies providing audit and certification of AIMS | 2025 | Requirements for 42001 certification bodies | + +### 3.1 Combinations Often Referred to Together in Certification Audits + +| 42001 clause | SC 42 standard referred to together | Use | +|----------|------------------------|------| +| §5.2 AI policy | ISO/IEC 38507 (governance) | Stating board responsibilities explicitly | +| §6.1.2 Risk assessment | ISO/IEC 23894 | Standardizing risk identification, analysis, and evaluation procedures | +| §6.1.4 Impact assessment | **ISO/IEC 42005** | Eight assessment areas, identification of interested parties | +| §7.3 Awareness and competence | ISO/IEC 22989 | Consistent training materials through common terminology | +| §8.5 Life cycle | **ISO/IEC 5338** | Consistency of processes at each stage | +| §9.1 Performance evaluation | ISO/IEC TR 24368 | Measuring ethical impact | + +### 3.2 Combining Open Source Compliance with the SC 42 Family + +| Area | Open source standard | SC 42 standard | Integrated use | +|------|--------------|-----------|----------| +| **AI SBOM** | ISO/IEC 5230 §3.3.1 | ISO/IEC 42003 (in development) | SPDX 3.0 AI Profile + CycloneDX 1.6 ML-BOM | +| **Vulnerability management** | ISO/IEC 18974 §4.3.2 | ISO/IEC 23894 | Integrating OSS CVEs into the AI risk assessment | +| **Impact assessment** | - | **ISO/IEC 42005:2025** | Including the impact of training data sources and licenses in the assessment areas | +| **AI life cycle** | ISO/IEC 5230 §3.3 | **ISO/IEC 5338:2023** | Mandatory review of OSS components at each stage | + +{{% alert title="Priority for adopting the SC 42 family" color="success" %}} +Organizations preparing for ISO 42001 certification are recommended to review the SC 42 family standards in the following order: +`(1) 22989 (terminology) → (2) 23894 (risk) → (3) 42005 (impact assessment) → (4) 5338 (life cycle) → (5) 42003 (implementation guidance, once published)`. +Using these five standards together with 42001 makes it possible to meet the level of specificity that certification auditors require without difficulty. +{{% /alert %}} + +--- + +## 4. Comparison of Requirement Formats + +The three standards express requirements in fundamentally different ways. + +### ISO/IEC 5230 and 18974 Approach: Numbered Verification Materials + +Each clause specifies by number the verification materials (Verification Material) the company must provide. + +``` +§3.1.1 Policy + Verification materials: + - 3.1.1.1 A documented open source policy + - 3.1.1.2 A documented procedure to communicate the policy +``` + +If the verification material exists, the requirement is ✅ met; if not, it is ❌ not met, so the judgment is clear. + +### ISO/IEC 42001 Approach: Management System "Shall" Requirements + +Each clause presents **requirements as principles** in the form "the organization shall ...", +and the organization decides, according to its context, which documents or records will satisfy them. + +``` +§5.2 AI policy + "Top management shall establish an AI policy. + The AI policy shall be appropriate to the purpose of the organization..." +``` + +For this reason, ISO 42001 is difficult to self-certify with a simple checklist as with ISO 5230/18974, +and an internal gap analysis or an audit by an external certification body is needed. + +--- + +## 5. Comparison of Self-Certification Methods + +| Item | ISO/IEC 5230 | ISO/IEC 18974 | ISO/IEC 42001 | +|-----------|-------------|--------------|--------------| +| **Official self-certification tool** | OpenChain online checklist | OpenChain online checklist | None | +| **Self-certification cost** | Free | Free | Free (internal effort required) | +| **Basis for self-certification** | Declaration of checklist completion | Declaration of checklist completion | Self-declaration after an internal gap analysis | +| **Independent assessment** | OpenChain partners | OpenChain partners | Consulting firms | +| **Third-party certification** | OpenChain-accredited bodies | OpenChain-accredited bodies | ISO certification bodies such as BSI and TÜV SÜD | +| **Certification renewal** | 18 months | 18 months | According to the contract with the ISO certification body | + +--- + +## 6. Comparison of Open Source Relevance + +| ISO 42001 clause | Open source intersection | Corresponding ISO 5230 | Corresponding ISO 18974 | +|----------------|-------------------|--------------|---------------| +| §5.2 AI policy | Including OSS use principles in the AI policy | §3.1.1 Policy | §4.1.1 Policy | +| §6.1.2 AI risk assessment | OSS license and vulnerability risks | - | §4.3.2 Security assurance | +| §6.1.4 AI impact assessment | Impact analysis of OSS components | - | §4.1.5 Standard practice | +| §7.2 Competence | OSS compliance competence | §3.1.2 Competence | §4.1.2 Competence | +| §7.5 Documentation | AI SBOM | §3.3.1 SBOM | §4.3.1 SBOM | +| §8.5 AI life cycle | OSS framework licenses | §3.3 Content review | §4.3 Content review | +| §8.6 AI data | Open dataset licenses | §3.3.2 Licenses | - | +| §8.8 External AI procurement | Supply chain verification of OSS models | §3.3 Content review | §4.3.2 Security assurance | +| §9.1 Performance evaluation | OSS compliance metrics | §3.6 Conformance | §4.4 Conformance | + +--- + +## 7. Which Standard Should You Start With? + +{{% pageinfo %}} + +**If you have no open source compliance framework** + +Start with **ISO/IEC 5230**. +Build the foundation of license management, SBOM, policy, and training. + +--- + +**If you have an ISO 5230 framework and also develop AI** + +Review **ISO/IEC 18974 + ISO/IEC 42001** in parallel. + +- ISO 18974: Strengthen management of vulnerabilities in open source used in AI systems +- ISO 42001: Establish governance for AI systems as a whole (including the requirements that intersect with open source) + +The two standards complement each other, so pursuing them at the same time reduces duplicated work. + +--- + +**If you are a company that develops and provides AI systems as services** + +Check the open source intersection requirements of **ISO/IEC 42001** first. +The [Operation section](../4-operation/) of this guide describes the items you need to check right away in AI systems. + +--- + +**If you want to follow the trend toward mandatory AI SBOMs** + +Watch developments in **ISO/IEC 42003**. +The OpenChain AI Work Group is working to have the AI SBOM compliance guide reflected in ISO 42003. +The EU Cyber Resilience Act (CRA) is also being discussed in a direction that requires AI SBOMs as a means of transparency for AI systems. + +{{% /pageinfo %}} + +--- + +## 8. Common Foundation When Operating the Three Standards Together + +Common elements that can serve as a single foundation when complying with the three standards at the same time: + +| Common foundation element | ISO 5230 | ISO 18974 | ISO 42001 | +|----------------|:--------:|:---------:|:---------:| +| Open source policy | ✅ §3.1.1 | ✅ §4.1.1 | ✅ §5.2 (included in the AI policy) | +| Roles and responsibilities document | ✅ §3.1.2 | ✅ §4.1.2 | ✅ §5.3 | +| Competence and training records | ✅ §3.1.2 | ✅ §4.1.2 | ✅ §7.2 | +| SBOM / AI SBOM | ✅ §3.3.1 | ✅ §4.3.1 | ✅ §7.5 | +| Risk assessment | - | ✅ §4.3.2 | ✅ §6.1.2 | +| Responding to external inquiries | ✅ §3.2.1 | ✅ §4.2.1 | ✅ §8.7 (feedback channel) | +| Confirming and renewing conformance | ✅ §3.6 | ✅ §4.4 | ✅ §9, §10 | diff --git a/content/ko/guide/iso42001_guide/1-context-leadership/_index.md b/content/ko/guide/iso42001_guide/1-context-leadership/_index.md index ead6e9388a..fa8ba6ee28 100644 --- a/content/ko/guide/iso42001_guide/1-context-leadership/_index.md +++ b/content/ko/guide/iso42001_guide/1-context-leadership/_index.md @@ -44,14 +44,14 @@ AI 관리 시스템의 목적에 영향을 미치는 내·외부 이슈를 파 | **EU AI Act §11 + Annex IV** | 2027 (고위험) | 기술 문서 작성 의무 | AI SBOM이 기술 문서의 핵심 요소 | | **한국 AI 기본법** | 2026-01 시행 | 고영향 AI 분류·영향 평가·AI 시스템 표시 의무·학습 데이터 출처 공개 | AI SBOM·모델 카드 작성 의무 | | **US Copyright Office AI 가이드** | 2023-03 발행, 2024 갱신 | 완전 AI 생성물 = 인간 저작자성 부재(회사 저작권 등록 불가). 인간 검토·수정 비율에 따라 저작권 인정 | AI 코딩 도구 출력물의 저작권 귀속 처리 | -| **US EO 14110** (2023) | 시행 중 | 연방 AI 시스템 SBOM 요건·이중 사용 AI 모델 보고 | AI SBOM 표준화·NIST AI RMF 채택 | +| **US EO 14110** (2023) | 2025-01 폐지 (EO 14179로 대체) | 연방 AI 시스템 SBOM 요건·이중 사용 AI 모델 보고 | AI SBOM 표준화·NIST AI RMF 채택 | | **NIST AI RMF 2.0** | 2024 발행 | AI 리스크 관리 표준 프레임워크 (GOVERN·MAP·MEASURE·MANAGE) | ISO/IEC 42001과 상호 보완 | | **OSAID 1.0** (OSI, 2024-10) | 발효 | "오픈소스 AI 모델" 정의 — 데이터·코드·가중치 3요소 공개 | "Open Weight"(Llama·Gemma)는 OSAID 미충족 → 표 분류 정확성 | | **ISO/IEC 42005** (2025) | 발행 | AI 시스템 영향 평가 표준 | §6.1.4 영향 평가의 구체 기법 | | **ISO/IEC 23894** (2023) | 발행 | AI 리스크 관리 표준 | §6.1.2 리스크 평가 보완 | | **ISO/IEC 5338** (2023) | 발행 | AI 시스템 라이프사이클 표준 | §8.5 생애주기 단계 정의 | -| **ISO/IEC 42003** | DIS (2026-05 기준) | AI 경영시스템 인증 가이드 | OpenChain AI WG 입력 추진 중 | -| **ISO/IEC 42006** | 2026 발효 | AI 인증기관 요구사항 | ISO 17021-1 기반, 3자 인증 신뢰성 | +| **ISO/IEC 42003** | 개발 중 (신규 프로젝트 단계) | AI 경영시스템 구현 가이드 | OpenChain AI WG 입력 추진 중 | +| **ISO/IEC 42006** | 발행 (2025-07) | AI 인증기관 요구사항 | ISO 17021-1 기반, 3자 인증 신뢰성 | {{% alert title="조직 맥락 분석 시 활용" color="info" %}} 이 매트릭스를 조직 맥락(§4.1) 분석 시 외부 이슈로 매핑한다. 본 가이드의 다른 페이지 diff --git a/content/ko/guide/iso42001_guide/2-planning/_index.md b/content/ko/guide/iso42001_guide/2-planning/_index.md index 5c124b1012..02b52a993f 100644 --- a/content/ko/guide/iso42001_guide/2-planning/_index.md +++ b/content/ko/guide/iso42001_guide/2-planning/_index.md @@ -71,7 +71,7 @@ AI 시스템이 개인, 그룹, 사회에 미칠 수 있는 영향을 평가할 오픈소스 컴포넌트로 인한 영향도 포함한다. {{% alert title="ISO/IEC 42005:2025 — AI System Impact Assessment" color="info" %}} -ISO/IEC 42005:2025(2026-01 발행)는 §6.1.4 AI 시스템 영향 평가의 **구체적 수행 방법**을 +ISO/IEC 42005:2025(2025-05 발행)는 §6.1.4 AI 시스템 영향 평가의 **구체적 수행 방법**을 제공하는 SC 42 패밀리 표준이다. ISO 42001은 "영향 평가를 수행해야 한다(shall)"고 원칙만 명시하지만, 42005는 평가 범위, 시점, 이해관계자, 문서 형식까지 안내한다. diff --git a/content/ko/guide/iso42001_guide/compare/_index.md b/content/ko/guide/iso42001_guide/compare/_index.md index acc1a864ad..be9a4b6d4e 100644 --- a/content/ko/guide/iso42001_guide/compare/_index.md +++ b/content/ko/guide/iso42001_guide/compare/_index.md @@ -72,7 +72,7 @@ AI 표준 패밀리의 핵심이다. 인증 심사·실무 적용 시 다음 표 | **ISO/IEC 42001:2023** | AI Management System (AIMS) | 2023 | **인증 가능한 핵심 표준** | | **ISO/IEC 42003** | AIMS implementation guidance | 개발 중 | 42001 구현 가이드 (AI SBOM 포함 추진) | | **ISO/IEC 42005:2025** | AI System Impact Assessment | 2025 | 42001 §6.1.4 영향 평가 구체 방법 | -| **ISO/IEC 42006** | Bodies providing audit and certification of AIMS | 개발 중 | 42001 인증기관 요건 | +| **ISO/IEC 42006** | Bodies providing audit and certification of AIMS | 2025 | 42001 인증기관 요건 | ### 3.1 인증 심사에서 자주 함께 참조되는 조합 diff --git a/content/ko/guide/ai-sbom_guide/5-tools/1-aibom-generator/images/input-form.png b/content/shared/guide/ai-sbom_guide/5-tools/1-aibom-generator/images/input-form.png similarity index 100% rename from content/ko/guide/ai-sbom_guide/5-tools/1-aibom-generator/images/input-form.png rename to content/shared/guide/ai-sbom_guide/5-tools/1-aibom-generator/images/input-form.png diff --git a/content/ko/guide/ai-sbom_guide/5-tools/1-aibom-generator/images/result-view.png b/content/shared/guide/ai-sbom_guide/5-tools/1-aibom-generator/images/result-view.png similarity index 100% rename from content/ko/guide/ai-sbom_guide/5-tools/1-aibom-generator/images/result-view.png rename to content/shared/guide/ai-sbom_guide/5-tools/1-aibom-generator/images/result-view.png diff --git a/content/ko/guide/finance-oss-guide/5-manage/30th-s1-claude-code.png b/content/shared/guide/finance-oss-guide/5-manage/30th-s1-claude-code.png similarity index 100% rename from content/ko/guide/finance-oss-guide/5-manage/30th-s1-claude-code.png rename to content/shared/guide/finance-oss-guide/5-manage/30th-s1-claude-code.png diff --git a/content/ko/guide/finance-oss-guide/5-manage/dt-projects.png b/content/shared/guide/finance-oss-guide/5-manage/dt-projects.png similarity index 100% rename from content/ko/guide/finance-oss-guide/5-manage/dt-projects.png rename to content/shared/guide/finance-oss-guide/5-manage/dt-projects.png diff --git a/content/ko/guide/finance-oss-guide/artifacts/3-audit-evidence/30th-s3-key-questions.png b/content/shared/guide/finance-oss-guide/artifacts/3-audit-evidence/30th-s3-key-questions.png similarity index 100% rename from content/ko/guide/finance-oss-guide/artifacts/3-audit-evidence/30th-s3-key-questions.png rename to content/shared/guide/finance-oss-guide/artifacts/3-audit-evidence/30th-s3-key-questions.png diff --git a/content/ko/guide/finance-oss-guide/artifacts/4-tool-recipe/dt-dashboard.png b/content/shared/guide/finance-oss-guide/artifacts/4-tool-recipe/dt-dashboard.png similarity index 100% rename from content/ko/guide/finance-oss-guide/artifacts/4-tool-recipe/dt-dashboard.png rename to content/shared/guide/finance-oss-guide/artifacts/4-tool-recipe/dt-dashboard.png