Telemetry: a persistent machine_id makes the data pseudonymous, and default-on collection lacks prior consent for EU users (GDPR / ePrivacy)

[Gotman08]

Summary

Telemetry is described as anonymous, but every payload carries machine_id, a random UUID stored in ~/.codegraph/telemetry.json that never expires. Because it links all events from the same machine over time, the data is pseudonymous rather than anonymous. Under the GDPR, pseudonymous data is still personal data. Combined with the default-on setting, I believe the current design is unlikely to be compliant for users in the EU, and I'd like to suggest some changes.

Context: CodeGraph 1.6.0 on Windows 11 (npm install). I never ran the interactive installer, so no toggle was shown. The one-line notice appeared during my first codegraph index, in the same run as the first POST, so there was no opportunity to refuse before data left the machine.

Details

1. Pseudonymous, not anonymous. A stable identifier that lets you recognise the same machine across sends is pseudonymisation (GDPR Art. 4(5), Recital 26), and the data stays within the scope of the GDPR. The words "anonymous usage stats" in TELEMETRY.md, the CLI notice and the telemetry command output are therefore misleading for EU users. The client IP is also exposed to the hosting provider when the request is made, even though the worker itself never reads it.

2. Consent before storing an identifier on the device. Writing an identifier to the user's machine and reading it back for analytics falls under Art. 5(3) of the ePrivacy Directive (in France, Article 82 of the Loi Informatique et Libertés). It requires prior consent unless the storage is strictly necessary for the service the user asked for, and usage statistics are not. The EDPB Guidelines 2/2023 on the technical scope of Art. 5(3) make clear this is not limited to browser cookies. A default-on toggle, or a notice printed when the first send happens, is not valid consent: consent must be a clear affirmative act (CJEU C-673/17, Planet49).

3. Retention. Raw events are purged after 90 days, but according to telemetry-worker/migrations/0001_init.sql, machine_days and machine_first_seen keep the machine ID and its active days with no limit. For comparison, the French regulator (CNIL) only exempts audience measurement from consent when the identifier lasts at most 13 months and the data is kept at most 25 months. The current design does not meet those conditions either.

Suggestions

• Make telemetry opt-in, or at least keep it off in non-interactive runs until the user has explicitly made a choice.
• If a persistent ID is kept, rotate it (for example every 13 months at most) and put a retention limit on machine_days and machine_first_seen. Alternatively, drop the persistent ID in favour of a per-day random value.
• Describe the data as pseudonymous in TELEMETRY.md, the notice and the CLI messages. Also name the data controller and give a contact for access and erasure requests (a deletion request can only be matched through the machine ID).

I'm not a lawyer, and I'm happy to be corrected on any of these points. Thanks for the detailed TELEMETRY.md and for publishing the worker code, which made all of this easy to check.