On 9 September 2026, Ant International, Visa and Mastercard announced a shared Know Your Agent (KYA) interoperability framework so that an AI agent verified by one of them does not have to be verified again by the other two. Know Your Agent is the practice of identifying an AI agent, tying it to an accountable operator, and monitoring it before and while it spends money. It is the first time the two largest card networks have agreed a common approach to agent identity. It is also, for now, an agreement in principle: no technical specification, governing body or rollout date has been published.
What Is the Know Your Agent Framework?
The framework is a commitment to make agent identity portable across payment networks. Today each network runs its own agent verification protocol, and an agent platform that wants to transact across all of them has to onboard separately with each. The three companies say the new framework will streamline agent onboarding and identification across card networks, digital wallets, agent platforms and marketplaces.
It rests on three pillars, as described in the announcement and reported by Yahoo Finance:
| Pillar | What it is meant to do |
|---|---|
| Cross-network operator traceability | Link every agent to a validated operator, cardholder or business |
| Shared certification requirements | Assess an agent's security and behavioural compliance against one common bar |
| Continuous transaction monitoring | Use identity and transaction signals to keep oversight after onboarding |
Jiang-Ming Yang, Chief Innovation Officer at Ant International, put the practical promise plainly, as quoted by Forkast: "If an agent registers with Ant, they don't need to register again with Visa, Mastercard."
Which Protocols Does KYA Bridge?
KYA does not replace anything. Ant's inclusion matters because it brings the mobile-wallet side of payments, which neither card network runs. The framework sits above three identity protocols that each company shipped separately over the past year.
| Protocol | Owner | Launched | What it does |
|---|---|---|---|
| Trusted Agent Protocol (TAP) | Visa, co-developed with Cloudflare | October 2025 | Cryptographic check that an agent is who it claims to be, before payment |
| Verifiable Intent | Mastercard, co-developed with Google | March 2026 | Tamper-resistant record linking the user's identity, their instruction, and the agent's action |
| Agentic Mobile Protocol (AMP) | Ant International | April 2026 | Open-sourced protocol for agent payments through mobile wallets |
The three answer different questions. TAP asks whether the agent is genuine. Verifiable Intent asks whether the human actually authorised this purchase; it is open source, aligned with Google's AP2 and Universal Commerce Protocol, and draws on FIDO Alliance, EMVCo, IETF and W3C standards. Ant's AMP carries agent payments into the wallet economy rather than the card rails. KYA's job is to let an identity established under any one of them be recognised by the others.
How Is Know Your Agent Different From Know Your Customer?
Know Your Customer verifies a person or business before a financial institution serves them. Know Your Agent verifies software acting for that person or business, and it has to answer a harder set of questions.
| Know Your Customer (KYC) | Know Your Agent (KYA) | |
|---|---|---|
| Who is verified | A person or legal entity | An AI agent and the operator accountable for it |
| When | At onboarding, with periodic review | At onboarding and continuously, on every transaction |
| Core question | Is this customer who they claim to be? | Is this agent genuine, who stands behind it, and is it acting within what it was authorised to do? |
| Basis | Regulatory obligation | Industry framework, as announced |
The difference that matters most is continuity. A verified human customer stays the same person. A verified agent can be updated, re-prompted or compromised between one transaction and the next, which is why the framework's third pillar is continuous monitoring rather than a one-time check. KYA as announced is an agreement between companies, not a regulatory requirement, and that shapes how quickly and how uniformly it can be enforced.
Why Agent Identity Became the Bottleneck
Each network built its own agent verification protocol precisely because agents cannot be trusted by default. The problem KYA addresses is that those protocols do not recognise one another.
Card fraud controls were built for humans. They read device fingerprints, typing cadence, IP history and purchase patterns, and an AI agent looks anomalous on almost every one of those signals. A network that cannot tell a legitimate agent from a scripted bot has two options: decline more agent transactions, which kills adoption, or approve more, which invites fraud. Each network built its own way out. The result was a fragmented identity layer in which a single agent platform could hold a verified status on Visa and be a stranger on Mastercard.
The executives framed it in the same terms. Rubail Birwadker, SVP at Visa, said: "Without trusted identity and explicit permissioning, AI agents cannot participate in commerce at scale." Pablo Fourez, Mastercard's Chief Digital Officer, said: "Interoperability across Know-Your-Agent frameworks is essential to making agentic commerce work at scale." Both quotes are from the Forkast report.
Demand is the other half of the problem. The companies cite projections that AI agents will orchestrate US$3 trillion to US$5 trillion of global consumer commerce by 2030, a range that originates in McKinsey's October 2025 agentic commerce research. It is a forecast rather than a measurement, and the announcing companies have an obvious interest in its upper end, but it explains why all three want the identity question settled early. And consumers are not yet convinced: Visa's own Trust Index, released the same day, found only 23% of US consumers trust generative AI to handle payments on their behalf. Identity that follows an agent across networks is the precondition for moving that number.
What Has Not Been Decided
The announcement is notable for what it leaves out, and those omissions define how much weight it can bear today.
- No specification. There is no published document describing how a verified status on one network is expressed, transmitted or checked by another.
- No governing body. Nobody has been named to set the shared certification bar, audit against it, or revoke an agent's status.
- No timeline, pilot or volume. No launch date, participating merchants or transaction counts have been disclosed.
- No liability model. The announcement does not say whether a merchant that accepts a KYA-verified agent is protected if that agent turns out to be fraudulent.
- No bridge between rails. Visa and Mastercard verify agents on card networks; Ant's protocol was built for mobile wallets. How one identity is honoured across two different payment architectures has not been described.
The PYMNTS report describes the work as exploratory. Forkast's headline puts it more directly: the parties have agreed, and now comes the hard part.
Revocation is the least discussed of these gaps. A portable identity is only safe if its withdrawal is portable too: an agent struck off by one network for fraudulent behaviour has to lose its standing on the others at the same moment, or the framework simply lets a bad actor choose the most lenient door. The announcement describes how agents join; it does not yet describe how they are removed.
The governance gap is the one to watch. A shared certification bar is only as strong as whoever enforces it, and three competitors with proprietary protocols have commercial reasons to keep their own onboarding as the preferred route. The x402 protocol resolved the same tension by moving into a neutral foundation. KYA has not yet said how it will resolve it.
Is This the Card Networks Closing Ranks?
The contrarian reading is that KYA is less about interoperability than about jurisdiction. Agent identity is the gate every agent payment must pass through, and whoever defines "verified" decides which agents can transact. By agreeing a common standard among themselves, Visa, Mastercard and Ant position the card networks and the largest wallet operator as the natural issuers of agent identity, ahead of the AI platforms that build the agents and the crypto rails that route around cards.
That reading does not make the framework bad for merchants. A single verification that works across networks is strictly better than three separate ones, and it lowers the cost for a smaller agent platform to reach every network. But it does mean that the trust layer of agentic commerce is settling inside the existing payments incumbents, not beside them. The absence of Google, OpenAI, Stripe or Coinbase from the announcement is worth noting, even though Google co-developed Verifiable Intent.
What Merchants and Agent Builders Should Do Now
Nothing in the framework requires action this quarter, because nothing in it is yet implementable. The sequence that follows from the three pillars is still clear.
The first step is to map which of the three underlying protocols your payment stack already touches. A merchant on a processor that supports TAP or Verifiable Intent is already inside the perimeter KYA is meant to widen; one that supports neither will need their processor to move first.
The second is operator traceability, the pillar most likely to create work. If agents must be linked to a validated operator, merchants and agent platforms should expect to hold and pass through a record of which operator stands behind each agent session. Logging that relationship now costs little and turns a future certification requirement into a configuration change rather than a rebuild.
The third is to keep checkout decisions reversible, and to treat KYA as preparation rather than a commercial benefit until liability is defined. A verification that does not shift fraud liability away from the merchant changes little about the merchant's risk. Until a specification exists, betting a checkout integration on any one network's agent identity scheme is a bet on the scheme, not the standard.
