Visa, Mastercard and Ant are building an ID card for shopping agents — and leaving the liability question open

What happened

The three biggest players in agentic payments have started to agree on a common form of identification. On 9 September 2026 Ant International, Mastercard and Visa announced that they are working on an interoperable Know Your Agent framework — KYA for short. The aim is for card networks, wallet ecosystems, agent platforms and marketplaces to be able to onboard and identify AI agents according to shared principles, without any provider giving up its own checks. In practice that means an agent which has identified itself to one payment provider should not have to repeat the process with the others. Each of the three companies brings its own protocol: Visa’s Trusted Agent Protocol, Mastercard’s Verifiable Intent and Ant International’s Agentic Mobile Protocol. The coordination runs through BuildFin.ai, an industry platform convened by the Monetary Authority of Singapore. According to TechNode Global, liability, the revocation of an agent’s authorisation, data sharing and the handling of disputes are explicitly still open.

Who it affects

Online merchants who will receive traffic from shopping agents over the next two years, whether they want it or not. Shops with high basket sizes, configurable products or tight return windows are particularly exposed — there, the difference between “the customer ordered” and “the agent ordered” is hard cash. For payment service providers, payment network operators (Netzbetreiber) and shop platforms it is a signal that the schemes intend to regulate agent traffic rather than ignore it.

Assessment

Two competing card networks and the largest Asian wallet operator moving towards a common language is the grown-up response to a year full of individual announcements. Three proprietary protocols side by side would have meant, for merchants: integrate three times, read three rulebooks, argue three times. A shared form of identification for agents is therefore the right construction site — it answers the question of who is actually knocking, and it stops every marketplace inventing its own approval procedure.

What it does not answer is the question that counts in retail. A verified agent is not an authorised purchase. What happens if the agent orders within the scope of its credentials but outside what the customer actually wanted? Who carries the chargeback when the customer says they never ordered this — the cardholder, the agent platform, the issuer or, once again, the merchant? Those are precisely the points that, according to the reporting, are still on the open list, and experience says: as long as the burden of proof is not explicitly shifted in the rulebook, it stays where it always sits in the card world. So the right stance for merchants is neither euphoria nor rejection, but bookkeeping. Anyone who cannot distinguish today which orders came from an agent will not be able to prove in two years that their own chargeback rate is an agent problem and not a merchant problem. That distinction is not an IT project, it is a data field.

What to do now

1. Ask your payment service provider and shop vendor whether agentic orders are delivered flagged as such, and take that attribute into your own reporting.
2. Track the chargeback rate separately for agent and direct orders as soon as the flag is available — that is later the only solid basis for negotiation.
3. Before joining any agent programme, get it in writing who is liable in a dispute and under what conditions an agent’s authorisation is revoked. If the answer is missing, participation is a test, not a sales channel.