Verifying AI agent identity at checkout
As autonomous AI agents start acting on behalf of users — placing orders, signing contracts, moving funds — merchants and platforms need a way to verify two things at once: which agent took the action, and what it actually decided. AIEF execution artifacts give you both, without trusting the agent runtime that produced them.
The problem
A checkout flow that accepts requests from AI agents has the same risk surface as one that accepts requests from anonymous scripts, but worse: the agent's decision is fluent, plausible, and after the fact almost impossible to reconstruct from logs alone. If a transaction is later disputed — by the user, the issuer, or a regulator — “the agent decided to buy it” is not evidence.
What AIEF gives you
AIEF defines an execution artifact: a structured, tamper-evident record of a single AI decision. The artifact captures the inputs the agent acted on, the issuing system's identity, the recorded output, and the integrity scheme used to seal it. Any party with the artifact can verify it deterministically, using public rules.
For agent commerce, the artifact answers the two checkout questions: which agent (issuer identity in the artifact header) and what it decided (the protected output fields).
End-to-end pattern
- Require an artifact alongside the agent request. When the agent posts a checkout intent, require an AIEF artifact at Level 2 or Level 3 conformance. Level 1 is fine for development; Level 2 and above are appropriate for production evidence.
- Verify the artifact before charging. Run the artifact through a compliant AIEF verifier. Reject on any non-success reason code. Verification is deterministic — the same artifact and rules always produce the same verdict.
- Bind the artifact to the order. Store the artifact (or its hash and retrieval URL) on the order record. If the transaction is later disputed, the artifact is the evidence.
- Check issuer identity against your allow-list. The artifact identifies the issuing system. Decide which agent issuers you accept — and at what spending limits — before the artifact ever arrives.
What this is not
AIEF verifies the integrity of the decision record, not the wisdom of the decision. An agent can still buy the wrong thing; AIEF makes sure you can prove exactly what it decided, on what inputs, under whose identity. Risk scoring, fraud detection, and spending limits are still your problem — AIEF just makes them enforceable against evidence instead of telemetry.
Next steps
- Read key concepts for the vocabulary used above.
- Read the current specification for the artifact schema and verifier contract.
- Review conformance levels to decide which level your checkout requires.