Skip to content
EU AI Act glossary · Obligations· Art. 12

Record-keeping (logging)

Article 12 requires high-risk AI systems to automatically record events (logs) over their lifetime, ensuring a level of traceability appropriate to the system's intended purpose.

Logging must be technically enabled by design, not added by the deployer afterwards. The logs must support identifying situations that may result in the system presenting a risk or undergoing a substantial modification, facilitate post-market monitoring, and enable deployer monitoring of operation. For high-risk remote biometric identification systems, Article 12(3) specifies the minimum content: the period of each use, the reference database against which input data was checked, the input data for which the search led to a match, and the identity of the persons involved in verifying the results. Deployers must keep the logs for a period appropriate to the intended purpose, of at least six months, unless other Union or national law provides otherwise.

Where Record-keeping (logging) sits in the Regulation
  • Art. 12
  • Art. 12(3)
  • Art. 19
  • Art. 26(6)
Who it binds

Record-keeping (logging): why it matters

Logs are what make every other obligation provable after the fact. Without them, a serious-incident report cannot be reconstructed, post-market monitoring has nothing to monitor, and a deployer facing an affected person's Article 86 request for an explanation has nothing to explain from. This is also one of the few requirements that is genuinely impossible to retrofit cheaply — a system that was not built to log the right events cannot start producing history for the period before someone noticed.

Worked example

Record-keeping (logging) in practice

A bank's loan-decision system logs, for every application, the model version, the input feature vector, the score, the threshold applied, whether a human overrode it and who. Eleven months later a rejected applicant exercises their right to an explanation. The bank can reconstruct the decision exactly. A competitor that logged only the final yes or no can reconstruct nothing, and has no defence to offer either the applicant or the regulator.

Correction

Record-keeping (logging): the mistake people make

Assuming ordinary application logs satisfy Article 12. Standard observability captures system health, not decision traceability — you need the inputs, model version, output and oversight actions tied to an identifiable event, retained for at least six months, and privacy-engineered so that retention does not create a GDPR problem of its own.

Questions

Record-keeping (logging): common questions

How long must AI Act logs be kept?
Deployers of high-risk systems must retain the automatically generated logs for a period appropriate to the intended purpose, of at least six months, unless other applicable Union or national law — such as data protection law — requires otherwise.
Who is responsible for logging, the provider or the deployer?
The provider must design the system to generate the logs automatically and, where it controls them, retain them. The deployer must keep the logs the system generates while it is under its control.
Do logging duties conflict with the GDPR?
They have to be reconciled rather than traded off. Retention should be purpose-limited to traceability, access controlled, and personal data minimised or pseudonymised where the traceability purpose can still be met.
Keep reading

Record-keeping (logging): terms that come up alongside it

See how Record-keeping (logging) applies to your systems

Run the free 2-minute EU AI Act checker — no sign-up.

Run the free check