What is an audit trail?
An audit trail is a durable, ordered record of actions taken by a system: who initiated them, what was read, what decision was made, and what changed.
For AI systems the bar is higher than for ordinary software, because the behaviour is not deterministic. Knowing that an action occurred is not enough; you need the context that produced it — the retrieved data, the policy decision, the exact arguments — or the record cannot answer why.
A trail that only proves something happened is a log. A trail that lets you reconstruct why is an audit.
Every run records identity, retrieved context, the policy decision, the provider request and the result, along with approvals and who granted them.
Related terms
- Human in the loop — A design where a person approves or rejects a consequential action before it executes.
- Row-level security — A database feature restricting which rows a query can see, enforced by the database rather than the application.
- Idempotency — A property where repeating an operation has the same effect as performing it once — essential when retries are possible.
See it working rather than described.
The product demo runs one agent across three systems and stops at the approval gate.