Definition
A standardised, machine-readable record of what an AI agent was authorised to do, who approved it, and what actually happened — signed and registered so it can be verified later. It is a standards proposal (an IETF draft) to give AI-agent actions auditable evidence of authority, approval and outcome.
Why it matters
When an autonomous agent does something wrong, the first question is 'who authorised this and can we prove what happened?'; OAR-style records make that answer possible instead of leaving it to inference after the fact.