Fetch.ai has released AEVS (Agent Execution Verification System), a tool that attaches cryptographic receipts to the actions AI agents take when calling external tools. The engineer behind it, Devendra Chauhan, described the goal directly: "AEVS creates cryptographic receipts for agent actions, making executions independently verifiable and auditable," "No black boxes. No 'trust me' logs." The design goal was adoption without rework: the goal was to make verifiable execution simple without forcing developers to redesign their stack. What's notable is the honesty about the tool's limits. Fetch.ai is explicit that AEVS is tamper-evident, not tamper-proof — the system detects modification after it happens, yet it does not prevent the modification in the first place. That's a real distinction, and it cuts against some secondary coverage: some secondary coverage describes AEVS as an on-chain tool, with CryptoBriefing framing it as receipts anchored on-chain — a claim Fetch.ai itself doesn't fully endorse. This lands in a busy month for the receipts idea. Separate projects are converging on similar architecture: signed, hash-chained records of individual tool calls that a third party can check without trusting the agent operator. One academic framework describes the pattern generally — an attestation layer that automatically generates a signed receipt whenever the agent makes an external call, with each receipt a compact, tamper-proof record appended to an immutable Provenance Log. Another, built for compliance auditing rather than crypto rails, ties receipts to enforceable dates: it frames the need around EU AI Act Article 12, enforceable August 2, 2026, where the question isn't 'did you check?' but 'can you prove you checked, and can I verify that proof without trusting you?' The pattern across all of these is the same: move from logs an operator controls to receipts a stranger can check. Fetch.ai's contribution isn't the idea — it's a working implementation shipped by a team with a track record in autonomous agent payments, tied to real infrastructure rather than a whitepaper. The gap that remains is the one Fetch.ai names itself: detection after the fact isn't prevention. A signed receipt tells you tampering happened; it doesn't stop an agent from acting on bad data before the receipt is checked. That's the honest state of agent attestation in mid-2026 — real cryptographic tooling, still one layer short of guarantees.