How does the receipt test work for AI agent vendor due diligence?

Ask the vendor for a verifiable execution receipt from a live run, then recompute its cryptographic commitment yourself. That is the whole test. A vendor with a real evidence chain hands you a receipt in minutes: tool, timestamp, inputs, outputs, and a SHA-256 commitment you can check with no account. A vendor with slides hands you slides. AER-1, published as an IETF Internet-Draft, is the open format that makes the test possible.

Want the proof yourself? Run one free call and get a real verifiable receipt, no signup.

Need more than the free tier? $0.99 unlocks unlimited runs for 24 hours, no signup, no subscription.

Updated October 7, 2026.

The procurement problem

Every AI agent vendor promises accuracy, security, and auditability. The RFP responses all read the same. The demos all work. Then the contract is signed, the agent touches production data, and the first incident reveals the truth: the vendor's auditability was a dashboard, the security review was a questionnaire, and nobody can prove what the agent actually did.

Due diligence for agents cannot be paperwork alone, because the risk is operational. The agent will act, spend, read, and write on your behalf. The question is not whether the vendor has a policy. The question is whether the vendor can prove, per action, what its system did.

The receipt test, step by step

  1. Ask for a live run. Have the vendor run their agent against a task you choose, in an environment you can observe.
  2. Ask for the receipts. For every tool call in that run, ask for a verifiable execution receipt: tool, calling agent, timestamp, inputs, outputs, and a cryptographic commitment.
  3. Recompute one commitment yourself. Take the canonical bytes from a receipt, hash them, and compare. You need no account and no vendor help.
  4. Ask for a gap. Ask what happens if a receipt is deleted or a call is missing. The right answer involves sequence numbers and visible gaps; the wrong answer is a shrug.
  5. Ask who minted it. If the agent wrote its own receipts, the evidence is circular. Receipts must be minted by the execution layer, outside the agent's reach.

A vendor that passes does it in minutes. A vendor that fails will explain why the test is unfair, which is itself the answer.

What the test reveals

  • Real evidence chains vs marketing. Passing vendors have per-call receipts wired into their runtime. Failing vendors have a logging roadmap.
  • Tamper posture. Receipts the agent cannot rewrite mean the vendor designed for adversarial review. Self-written logs mean they designed for demos.
  • Incident readiness. A vendor that can hand you receipts today can hand you evidence after an incident. One that cannot, cannot.
  • Cost honesty. Per-call receipts make per-call billing auditable, which is where vendor margin games hide.

Put it in the contract

The test is also a contract clause. Require verifiable execution receipts for every tool call touching your data, in an open format, retained for your audit window, delivered on demand. AER-1, published as the IETF Internet-Draft draft-zambo-aer1, is the open format to name: no vendor lock-in, public conformance vectors, anyone can verify.

The evidence standard to cite

Name AER-1 receipts in your vendor requirements: minted by the execution layer, SHA-256 commitment per call, independently verifiable. Read the format guide.

Run the test yourself first

Before you ask a vendor for a receipt, hold one yourself. The live demo mints a real AER-1 receipt for a real tool call, free, no signup. Recompute the commitment. That is the bar you are about to set.

Mint a receipt and learn the test in one minute.