Verifiable Receipt AI Agent
Last updated: 2026-10-03
In short
A verifiable receipt binds a saved execution result to integrity data so another reader can check whether the recorded bytes changed. It is not a log, and it is not proof that the answer was right.
What makes the receipt verifiable
The receipt identifies a recorded execution and exposes data a verifier can compare. If the saved bytes no longer match the commitment, the check fails. Read the AER-1 open draft for the receipt format and use the verifier to inspect a record.
Integrity is not correctness
Integrity answers whether the saved record changed. It does not determine whether the model reasoned well, whether the output is factually correct, or whether an external system accepted a side effect. Those claims need their own checks.
When to share a receipt
Share a public receipt when a reviewer needs to inspect a bounded tool call without relying on the original chat. Redact private context and check that the receipt exposes only the information appropriate for public review.
Frequently asked questions
Verifiable Receipt AI Agent
A verifiable receipt is a record of an execution with integrity data that another reader can check. It supports review of the saved record, not a claim that the model's answer was correct.
Is a verifiable receipt the same as a log?
No. A log is usually operational history. A receipt is a bounded record with a verification path and integrity material for independent checking.
What does a valid receipt establish?
It can establish that the saved record matches its integrity commitment. It does not establish that an unobserved outside event happened or that the answer was correct.