AI Agent Receipt Format
Last updated: 2026-10-03
In short
An AI agent receipt format defines the fields and byte rules a verifier needs to inspect a saved execution. The current reference is the open IETF draft draft-zambo-aer1-10, which can change as discussion continues.
AER-1 is an open IETF draft
AER-1 is an IETF draft, draft-zambo-aer1-10. It describes a portable receipt shape and verification process, and it remains open to revision. See the AER-1 page or its IETF Datatracker entry.
Short redacted shape
This partial example shows the kind of fields a verifier may inspect. It is deliberately redacted and incomplete, is not a live receipt, and cannot be used for verification.
{
"id": "[redacted]",
"created_at": "[redacted]",
"tool": { "name": "[redacted]" },
"canonical_bytes": "[redacted]",
"output_hash": "sha256:[redacted]"
}What the format does and does not establish
A format can make records easier to compare and verify. A successful integrity check means the inspected bytes match the record's commitment. It does not establish that an agent's answer was correct or that an outside action succeeded.
Frequently asked questions
AI Agent Receipt Format
AER-1, draft-zambo-aer1-10, is the open IETF draft referenced here for portable AI agent execution receipts. It is a draft that can change, not a finalized specification.
What is shown in the redacted example?
The short example sketches an identifier, timestamp, tool name, canonical bytes, and output hash. Its values are redacted and it is intentionally not a valid receipt.
Does the format prove an agent's answer is correct?
No. The receipt format supports integrity checks for recorded execution data. Correctness and outside outcomes need independent review.