Definition
AI decision logging is the structured recording of AI system decisions, inputs, outputs, actors, and system state so decisions can be traced, reviewed, and audited later.
Also available:ENFRsoonDEsoonITsoonESsoon
AI Governance Glossary
AI Decision Logging
AI decision logging gives organizations a machine-readable evidence trail for how an AI system behaved at a specific point in time. It is a core building block for auditability, accountability, incident review, and regulatory compliance. Decision logs complement application telemetry — where telemetry captures runtime performance, decision logs capture governance-relevant events: what the system decided, under which policy, with what rationale, and with what authority.
Why it matters
- •It creates a durable record of what the system did and when.
- •It helps teams reconstruct failures, overrides, and high-risk decisions.
- •It supports internal governance, external audits, and regulator requests.
- •Hash-chained decision records provide tamper evidence — altering any historical record breaks the chain.
Regulatory relevance
- •EU AI Act Article 12 requires automatic logging of events during the operation of high-risk AI systems, with a minimum six-month retention period.
- •EU AI Act Article 19 requires that providers maintain logs generated by their systems.
- •Decision logs support technical documentation, post-market monitoring, and accountability workflows under Articles 9, 11, and 17.
Implementation notes
- 1.Log timestamps, actor identity, model/version, decision outcome, relevant input/output references, and policy version.
- 2.Use SHA-256 hash chaining (RFC 8785 canonicalization) to create tamper-evident append-only chains.
- 3.Each record's record_hash becomes the previous_hash of the next record — genesis records have previous_hash: null.
- 4.Prefer structured event records over unstructured application logs. Design for retention, searchability, exportability, and chain verification.