Guidelines  ·  2026-11-20

IETF draft-malhotra-scitt-oar-00: 'Open Action Record (OAR): Authority, Approval and Outcome Evidence for AI Agent Actions'

GuidelinesMedium impactGlobal
New individual Internet-Draft submitted (last updated 2026-11-18), draft-malhotra-scitt-oar-00, 'Open Action Record (OAR): Authority, Approval and Outcome Evidence for AI Agent Actions.' It defines evidence that any agent-action profile can add to answer post-incident/dispute questions: an authority set binding digests of the outside authorizations an action relied on (with a re-check recorded at commit time), an approval binding tying a person's confirmation (passkey or approval service) to the exact action and what they were shown, salted commitments to affected state pre/post execution, an outcome confirmation made by reading the system of record, and three outcome classes a relying party can act on. Values are hidden behind salted commitments with field-by-field selective disclosure; a reference OAR format plus three verifiers agree on published test vectors.
This is the missing 'evidence for accountability' layer on top of the growing corpus of agent identity/delegated-authorization drafts. It addresses the practical case where an agent action caused harm and the relying party must prove whether the authorizations were still valid at commit, whether a named human approved that exact action, and whether the system-of-record state actually changed — the core of trustworthy agentic delegation and auditability.
Review the draft for agent action-evidence and audit use cases; consider comment/reference for agentic-authorization and SCITT registries.
IETF Datatracker — draft-malhotra-scitt-oar-00
See this in the live feed Explore related AI security and governance findings — updated every morning.
Open the feed →