An artificial intelligence agent audit trail preserves evidence about what an agent did and under which authority. Records can include the immutable identity, prompt and policy versions, requested targets, approvals, tool responses and resulting state changes.
Auditability supports incident investigation, compliance and evaluation. Logs should be tamper-resistant, access-controlled and connected across child processes so a reviewer can reconstruct one complete run.
