What happened
On 21 October 2026 a new Experimental individual Internet-Draft (draft-watts-agent-continuity-evidence-00) was published. It defines a signed JSON statement and verification model for observing computational-agent continuity across sessions, instances, and delegations, distinguishing model/runtime/agent/grant/execution identifiers, retaining predecessor evidence, and producing three-valued verification outcomes — explicitly separating record integrity from authorization.
Why it matters
Adds to the fast-growing IETF body of agentic-AI audit/evidence formats (alongside audit-trail, authorization-verdict, and rule-provenance drafts). It addresses a concrete blind spot: proving causal continuity of a delegated agent's execution for forensics and audit while avoiding conflation of authentic traces with authorized actions.
Action needed
Review for applicability to agent audit logging and incident-investigation pipelines; monitor for adoption in agent runtime implementations.