What is an Agent Audit Trail?

The defining property of an audit trail is that it cannot be quietly edited after the fact — entries are appended, often hash-chained so any tampering is detectable, and nothing is deleted to make history look cleaner. This matters because agents increasingly touch financial records, customer data, and regulated workflows, and a regulator, auditor, or incident-response team needs to reconstruct exactly what happened, in what order, under whose authorization, months or years later. A proper audit trail captures more than "the agent sent an email": it records the tenant, the session, the specific tool called, the permission scope it was granted under, and the outcome, so a reviewer can answer not just what happened but whether it was authorized to happen. This is a hard requirement across SOC 2, ISO 27001, and GDPR accountability obligations, and it's distinct from a user-facing activity log, which optimizes for readability over immutability.