AI agents can already research products, compare prices, and negotiate on a person's behalf. Handing one a payment decision is a different problem. A new long read on Finextra argues that the real question in agentic payments is not whether agents can transact, but whether they can be trusted to.
Most of the current work sits on the buyer's side. Coinbase, Stripe, Visa and Mastercard have each built frameworks for agent-initiated transactions. Account-to-account payments have no equivalent. Of the 40 members in the recently formed x402 foundation, built to develop internet-native payment rails, not one is a bank.
The piece places Know Your Agent (KYA) as the next entry in a taxonomy IPID already operates in, alongside know your customer, know your business, know your payee and know your transaction. Where those layers establish who a customer, business or payee is, KYA asks who is acting on a payment and under what authority.
The framework it proposes comes down to three questions: who is performing the action, what activity is being carried out, and what authority has been granted for it. Each needs an answer before a payment is initiated, not after a dispute is opened.
That pre-payment framing runs through IPID’s own work on payee verification. Confirming who receives a payment before it moves is the same instinct this piece applies to confirming who, or what, is initiating one — the same gap we've flagged before in how AI finance agents handle identity on the payer side.
IPID applies the same logic on the build side. The IPID Payment Agent turns invoices, messages or screenshots into verified payment instructions for banks and PSPs, running payee verification and Know Your Payee (KYP) checks before a payment is created, not after.
