How AI agents can pay under PSD2: the mandate model explained

PSD2 requires strong customer authentication for electronic payments: two of three factors among knowledge, possession and inherence. An AI agent has no fingerprint and owns no phone, so an agent can never satisfy SCA itself. For a while this looked like a wall for agent payments in Europe. It turned out to be a design constraint with a clean solution.

The mandate model

The solution is to move the authentication from the transaction to the mandate. A human sets the rules: monthly budget, single amount limit, allowed vendors, approval threshold. The human confirms this rule set once with a passkey, which is a textbook SCA event. Every later payment the agent makes inside those rules executes under that standing authorisation. Payments outside the rules do not get a softer check, they get a hard stop or a human approval request.

Precedent at network scale

This is not a boutique interpretation. When Visa took agent initiated payments live across Europe in July 2026 with more than 30 issuers, every transaction was tied to a verified user instruction through Visa Payment Passkeys, exactly the pattern of authenticating the human intent rather than asking a machine to fake a fingerprint. Mastercard's Agent Pay uses tokenised agent credentials with predefined limits under the same logic.

What builders should take from this

If you are building agent payments for the EU market, the architecture follows from the regulation: a deterministic policy engine that stores human authorised mandates, an SCA moment at mandate creation and modification, and an audit trail that can show, for any payment, which mandate authorised it and when the human signed. Get that chain right and PSD2 stops being the obstacle and becomes the moat.