One of the persistent gaps in agentic AI deployment has been the lack of a shared schema for logging what autonomous agents actually did. Every vendor and framework has invented its own audit format, which makes cross-system verification nearly impossible when a regulator, auditor, or incident responder needs to reconstruct an agent's decision chain. An IETF Internet-Draft filed in March 2026, draft-sharif-agent-audit-trail, attempts to close that gap. The draft's abstract states it specifies a standard logging format for autonomous AI agent systems, defining the Agent Audit Trail (AAT) as a JSON-based record structure with mandatory fields for agent identity, action classification, outcome tracking, and trust level reporting. Records are linked using tamper-evident hash chaining with SHA-256 per RFC 8785, with optional ECDSA signatures added for non-repudiation. The motivation section is direct about what's broken today. The document lists the current deficiencies as proprietary formats that vary across vendors making cross-system auditing impossible, no tamper-evidence allowing post-hoc modification of logs undetected, inconsistent action taxonomies preventing behavioral comparison across implementations, and no linkage between agent identity and logged actions, which makes attribution unreliable. Mechanically, each record is tied to a session identifier and a hash of the previous record, so the session_id plus prev_hash combination is what turns individual records into a verifiable chain, with every record cryptographically committing to the content of the record before it. The draft also defines trust levels ranging from L0 to L4, and it explicitly maps its fields to existing compliance regimes, citing EU AI Act Article 12, ISO/IEC 42001, SOC 2, and PCI DSS as regulatory references. The draft is not without gaps. Independent analysis of the spec notes it doesn't require pre-execution recording, meaning an agent could act before the corresponding audit record is committed, which weakens the guarantee that logs reflect decisions made in real time rather than reconstructed after the fact. Procedurally, this is still an early-stage document. As an Internet-Draft it carries no formal standing in the IETF process yet, and it expires September 29, 2026, at which point it either advances toward RFC status, gets revised, or lapses entirely. That timeline matters: it gives the agent infrastructure community roughly two months to weigh in before the draft's fate is decided one way or another. What's notable isn't that the problem is solved — hash-chained JSON logs don't resolve questions about whether an agent's underlying intent was legitimate — but that the audit-log interoperability gap has moved from a shared complaint into an actual draft specification with defined fields, regulatory citations, and an expiration clock.