What does it mean for an AI agent to be “verified” before conducting transactions?

Agent Verification Checklist (Per Transaction)

Use this checklist to determine whether an agent is “verified” for a specific transaction request:

• Identity binding — The agent’s credential/public key is registered and not revoked.
• Consent evidence — A valid consent receipt is present, signed, and unexpired.
• Consent scope match — The requested action is allowed by consent_scope.
• Constraint satisfaction — Amount, merchant/category, currency, time window, and geography are within bounds.
• Policy compliance — Issuer authorization rules approve the transaction for this agent + merchant + context.
• Anti-replay — The receipt nonce has not been used for prior approvals.
• Continuous verification — Verification runs every time (no “verified forever” shortcut).

If any check fails, return a binary YES/NO with a reason code so the agent and merchant can remediate safely.

Verified Agent Graph (Per Transaction)

This diagram shows the criteria that define whether an agent is verified for a specific transaction:

Verified Agent Failure Modes

Verification Requirements

A verified agent satisfies all of the following:

1. Identity binding — The agent has a cryptographic identity (public key or credential) that is registered and not revoked
2. Valid consent — The agent holds a consent receipt issued by the delegating user, with matching scope and unexpired
3. Policy compliance — The proposed transaction satisfies the issuer's authorization policy
4. Constraint satisfaction — Amount, merchant, currency, time window, and geographic restrictions are all within bounds
5. Non-replayed — The consent receipt nonce has not been used before

Identity Binding

Agent identity must be stable and verifiable. The agent's public key is registered with the verification layer. At each transaction, the verifier checks the key against the registry. If the key is unknown, revoked, or expired, the agent is not verified.

See authentication mechanisms for credential types.

Consent Scope

The consent receipt defines what the agent is allowed to do. Scope is an array of actions (e.g., ["pay", "book"]). If the agent tries to perform an action not in scope (e.g., "cancel"), the transaction is rejected.

Constraint Enforcement

Constraints are hard limits encoded in the consent receipt:

Continuous Verification

Verification is not a one-time check. Every transaction triggers a full verification pipeline. An agent that was verified for one transaction is not automatically verified for the next. This prevents scope creep and ensures consent is re-evaluated for each action.