Signed Authorization Verdict.
A signed authorization verdict is a short-lived, signed token recording whether one agent-initiated payment was within the spending policy the agent was given, so a seller or processor can check it offline before settling.
Agent payments raise a dispute question card networks have not had to answer at scale: was this purchase within what the agent was allowed to do? A signed verdict makes the answer a portable artifact. The party that checked the policy signs its decision; the party that settles keeps the token as evidence.
The useful design choices are the ones that remove ambiguity. Refusals are signed like approvals, review is explicitly not approval, and the token expires in five minutes, so a stale approval cannot be replayed later. Source: draft-dias-agent-payment-verdicts-00.
Our reading. Whether or not this draft goes anywhere, the pattern is sound: separate the policy decision from settlement and sign the decision. It sits next to consumer-authorized intent, which records what the user wanted, and agent transaction probability scores, which estimate risk. For JWT handling basics, see JWT explained.
Common questions
What is a signed authorization verdict for agent payments?
It is a signed JWT that records the outcome of checking one agent-initiated payment against a spending policy: allow, review, or a specific deny reason. Defined in the IETF individual draft draft-dias-agent-payment-verdicts-00 (October 9, 2026), it lets a seller or payment processor confirm the payment was in policy by checking the signature against the issuer's public keys.
Can a merchant settle on a review verdict?
No. In the draft, allow is the only code on which a relying party may settle. Review means the payment is above a threshold and awaits operator approval, and any other code should be treated as do not settle.
How does a signed verdict differ from a payment mandate?
A mandate states what the user authorized in advance, such as limits or a specific cart. A verdict is the result of checking one concrete payment against those mandates at a point in time. The verdict is short-lived evidence of a decision; the mandate is the standing authority.
Is the agent payment verdict draft a standard?
No. It is an individual Internet-Draft offered as input to IETF discussion, with no formal standing, and its author's company runs the implementation it describes. Treat it as one proposal among several for recording agent payment authority.
Related terms
In the guides
Authorization Patterns for Agentic Workflows: Delegation, Constraints, and Just-in-Time Permissions
AI agents need authorization models that handle delegated permissions, multi-step workflows, and least-privilege at machine speed. The patterns that work and the ones being invented.
JWT Explained: JSON Web Tokens, JWT Authentication, and the Pitfalls
JWT (JSON Web Token) is the dominant signed-token format for authentication and API authorization. How JWT tokens are structured, how JWT authentication works in OAuth 2.0 / OIDC, which algorithms to pin, and the recurring vulnerability classes that keep biting implementers.
PCI DSS 4.0 and CIAM: Identity Requirements for Payment Workloads
PCI DSS 4.0's Requirements 7, 8, and 10 directly constrain CIAM design for any system handling cardholder data. MFA, audit logs, role separation, and the gotchas that fail QSA audits.