EU AI Act Article 12, in plain English
What the record-keeping article asks for, and how a black-box recorder for agents fits in.
This is general information, not legal advice. Whether your system is "high-risk" depends on its use case; ask a lawyer.
What Article 12 says
High-risk AI systems must technically allow events to be recorded automatically (logs) over the lifetime of the system. The logs should make it possible to trace how the system functioned, help identify situations that may present a risk or a substantial modification, support post-market monitoring, and support monitoring of operation by deployers.
How long logs are kept
Article 19 requires providers to keep automatically generated logs under their control for at least six months, unless other EU or national law says otherwise. Deployers have a similar duty under Article 26.
When it applies
High-risk obligations were scheduled for August 2026, but the EU's proposed "Digital Omnibus" could delay some of them to December 2027. Treat the timing as uncertain and check the current status.
Does the law require tamper-evident logs?
No. Article 12 does not mandate a specific technique. But a log that anyone can quietly edit is weak evidence. Hash-chaining, where each entry includes a hash of the previous one, is a common engineering answer: change one entry and verification fails from that point on.
What this means for agent teams
- Log what each agent read and wrote, with which agent did it and when.
- Keep logs for at least six months, and make them easy to export.
- Be able to show the log has not been altered.
- Keep personal data in logs to what you need (GDPR still applies).
How LogicMem helps
Every LogicMem memory read and write is recorded in an append-only, hash-chained log. Our own production chain holds about 800,000 entries that verify. Any agent can run logicframe_audit_verify. LogicMem is a building block for your logging, not a compliance certificate.