Article 12 of the EU AI Act becomes enforceable on August 2, 2026. It requires providers of high-risk AI systems to keep logs proving compliance, and until recently there was no shared machine-readable format for that evidence. An IETF Internet-Draft, draft-marques-asqav-compliance-receipts, is trying to fill that gap by defining a compliance profile for signed action receipts that AI agents emit as they operate. The draft binds receipt fields directly to specific legal text. It describes itself as defining a profile that records machine-readable evidence of access-control decisions, binding receipt fields to Articles 12 and 26 of the EU AI Act and Article 17 of DORA, plus a list of US frameworks including the NIST AI RMF, the Colorado AI Act, and SEC Rule 17a-4. That's not a vague gesture at 'compliance' — it's a spec author trying to make one receipt format satisfy regulators on both sides of the Atlantic at once. Mechanically, a receipt is a signed JSON payload with a hash chain (a `previousReceiptHash` field), a decision and policy digest, and optional timestamp anchoring via RFC 3161 or OpenTimestamps for offline verifiability. The draft is explicit that anchors have retention rules: where the anchor type is OpenTimestamps, the upgrade from the initial calendar attestation to the Bitcoin block attestation must be completed within the 7-day profile-imposed bound. It also allows verifiers to reconcile receipts against software supply-chain evidence — a verifier may cross-check executable_hash against the executable identity declared in the SLSA attestation resolved through slsa_provenance_pointer — rather than treating any mismatch as an automatic failure. This sits inside a broader push toward independently verifiable agent evidence. A related Microsoft proposal frames the same problem plainly: compliance evidence needs to be something a third party can verify independently, without access to the operator's infrastructure. And a companion IETF draft on delegation receipts sketches a fuller chain — from a user-signed delegation receipt through hardware-measured model state to a chain-linked action log — so an auditor can verify each layer independently and confirm that a logged action was taken by the model the User authorized, acting within the defined scope. None of this is settled standard yet — it's drafts, expiring in December 2026 and January 2027 respectively, competing for adoption. But the timing is not incidental: with Article 12 enforcement landing this weekend, the receipt format is arriving right when regulators start actually asking for proof.