Use casesAgentic commerce
USE CASE · AGENTIC COMMERCE

Verifying an AI agent's authority to transact.

An agent that says "I am acting for this account" is a claim, not evidence. The merchant or bank on the other end still needs to know who this is for and whether it may act.

Use case

Authority to transact, verified rather than asserted.

The problem

When an AI agent places an order, authorises a payment or manages a subscription on a customer's behalf, the party on the other end of that transaction needs the same three facts it would need from a human: who the underlying customer is, whether whoever instructed the agent actually has the authority to act for the account it claims, and whether that account is real and belongs to that customer. An unverified assertion from the agent answers none of the three.

The application

The components that establish these facts for a human customer apply directly to a transaction where the party acting is software, and none of them are new. Identity already established for the account holder answers who the customer is without re-running a full identification for every automated transaction. Signing-authority checks, the same ones applied to a director signing an onboarding contract, establish whether whoever instructed the agent had the standing to act on the account, identified to AML standard. Account ownership resolves the same way it does for merchant bank verification, over open banking rails or through Verification of Payee (VOP), so the account the agent transacts against is confirmed to belong to the customer it claims. Ongoing monitoring, the same register, sanctions and status tracking applied elsewhere on the platform, keeps that standing current rather than assuming it holds forever.

Why it matters beyond agentic commerce

None of this is a new capability. It is the same identity, signing-authority, account-verification and monitoring components already doing this work for a person at a keyboard, applied to a transaction where the party acting is software. The trust question does not change: who is this for, does it have the standing to act, and does that standing still hold. What changes is only that the question now gets asked at machine speed, which is exactly why the answer needs to come from verified data rather than an unchecked assertion.