An AI audit trail is a structured, tamper-evident record of every governed event in an AI system — data accessed, policy decisions made, models and tools invoked, memory written, identities involved and outputs returned — captured automatically at runtime.
What is an AI audit trail?
Application logs record that a request occurred; an audit trail records why the system responded as it did. The distinction matters when an output is challenged, because the defensible answer is a reconstruction of the decision — what was retrieved, what policy allowed and denied, which model produced the text, what was redacted. Capturing this by construction rather than by instrumentation is what makes evidence available for a system nobody anticipated having to explain, and it is the raw material for ISO/IEC 42001 and EU AI Act record-keeping obligations.