Know Your Agent Could Become the KYC of AI Payments

Points of Focus

• Visa, Mastercard, and Ant International are developing an interoperable Know Your Agent framework for AI-driven payments.
• KYA could become a trust layer for agentic commerce, but it cannot replace KYC, authorization, or fraud controls.
• As AI agents gain spending authority, proving user intent and determining liability could become more important than simply identifying the agent.

For decades, banks and payment companies have built their defenses around a basic question: Who is the customer?

Now, as AI moves from recommending what people should buy to actually completing purchases for them, the payments industry has another identity problem to solve: Who, or what, is the agent?

Visa, Mastercard, and Ant International are trying to answer that question together. The three payments companies are collaborating on a Know Your Agent (KYA) interoperability framework designed to help payment networks, digital wallets, AI platforms, and merchants identify trusted AI agents across different ecosystems.

The ambition carries echoes of Know Your Customer (KYC). But while KYC establishes who is behind an account, KYA is emerging as a way of establishing which software is acting, who is responsible for it, and whether other participants in a transaction should trust it.

That distinction could become increasingly important as AI gains permission to spend real money.

Muse, Instinct, and your AI agent are about to need an ID. Visa, Mastercard and Alipay are writing the spec.

Muse, Grok Bot and Instinct all launched in the last few weeks, and all three will spend your money. Grok Bot, Muse and Instinct all get a single-use card from Stripe for… pic.twitter.com/kWnxNGOzWm

— Simon Taylor (@sytaylor) September 13, 2026

From shopping assistant to economic actor

Agentic commerce changes a familiar internet transaction in a subtle but important way. Instead of a customer browsing a website, choosing a product and pressing “buy,” an AI agent could search for a product, compare offers, and complete the purchase within parameters established by the customer.

Visa, Mastercard, and Ant International estimate that AI agents could orchestrate between $3 trillion and $5 trillion in global consumer commerce by 2030.

But delegating a purchase inserts another actor between the customer and merchant.

A merchant receiving an automated request needs to distinguish a legitimate shopping agent from an ordinary bot or malicious software. A payment provider needs to understand whether that agent is permitted to transact. And if something goes wrong, participants need a reliable trail showing who authorized what.

The payments industry has already been developing pieces of that infrastructure. Visa’s Trusted Agent Protocol uses agent-specific cryptographic signatures to help merchants distinguish recognized AI agents and understand their intentions. Mastercard has been developing agentic payment infrastructure around concepts including verifiable intent, while Ant International has developed its own agentic commerce technology.

KYA attempts to make trust portable across those different systems rather than forcing every participant to recognize agents independently.

Why KYA is starting to look like financial infrastructure

The collaboration focuses on three broad areas: tracing an agent to a validated operator, establishing common certification requirements, and continuously monitoring agents using identity and transaction signals.

Individual networks would still maintain their own verification, approval, and risk controls. The important change is that an agent verified in one environment could potentially carry recognizable trust signals when interacting with another.

That addresses a major scaling problem.

Without interoperability, an AI agent could face different registration, identity, and verification requirements every time it moves between wallets, card networks, merchants, or platforms. A common framework could make agent identity more portable without requiring the entire payments industry to operate a single centralized identity system.

The initiative is being developed through BuildFin.ai, convened by the Monetary Authority of Singapore, as financial institutions increasingly examine how autonomous AI can interact safely with financial infrastructure.

In that sense, KYA could eventually become to autonomous software what KYC became to customers: a foundational trust checkpoint before access to financial services is granted.

But the comparison has limits.

Knowing the agent does not mean knowing its intent

Nicolai Søndergaard, senior research analyst at Nansen, told AlphaWire that establishing an agent’s identity addresses only one part of the problem:

“Know Your Agent could become an important trust layer for agentic commerce, but it is not a substitute for KYC, authorization, or fraud controls. Agent identity can establish who sent a request and link it to an accountable operator. The more difficult questions concern what the user actually authorized, whether the agent stayed within scope, and who bears the loss when a validly authenticated system makes a mistake.”

That distinction becomes critical when an agent is legitimate — but its action is wrong.

Imagine a verified AI travel agent that has permission to arrange a holiday. Its identity might be cryptographically proven, but that alone does not establish whether its user authorized a 2,000-British-pound hotel, a 500-pound flight, or a non-refundable booking.

KYA can answer, “Which agent did this?” It cannot, by itself, answer, “Was this particular transaction what the human intended?”

That creates a much more complicated payments problem than identifying bots.

Søndergaard said a robust architecture will require several safeguards operating together:

“The strongest architecture will combine cryptographic identity with scoped payment tokens, signed intent, runtime policy enforcement and clear dispute rules. The industry challenge is less about proving that an agent exists (compared to a human or script) than about proving that its action was authorized, appropriately controlled and fairly attributable.”

This is why payment companies are developing controls beyond identity. Agent-specific credentials can identify the software, while tokenized payment credentials can restrict what it accesses. Spending limits can constrain how much it spends, and signed intent can preserve evidence of what the user instructed the agent to do.

Together, those mechanisms begin to create something closer to a chain of accountability.

The liability question could decide whether KYA works

The hardest question may arise after all those controls succeed.

Suppose an agent is authentic, the customer is genuine, and the payment credential is valid, but the AI misunderstands an instruction and buys the wrong product. Who pays?

Traditional payment systems have spent decades developing rules governing unauthorized transactions, chargebacks, merchant disputes, and fraud. Agentic commerce introduces a new category in which a transaction may be technically authorized but economically unintended.

That makes attribution as important as authentication.

KYA could establish that a recognized agent initiated the payment and connect that agent to an accountable operator. Other controls could establish what permissions the customer granted. Dispute frameworks would then have to determine responsibility when those systems disagree or fail.

The next KYC may need to answer more than “who?”

KYC became fundamental because financial institutions could not safely provide services without understanding the humans and businesses using them. Agentic commerce may create a parallel requirement for software.

Yet KYA faces a harder test.

The payments industry will need to establish not only which agent initiated a transaction but also who authorized it, what authority was delegated, whether that authority was exceeded, and who carries the loss when an autonomous decision goes wrong.

Visa, Mastercard, and Ant International’s interoperability push suggests those questions are moving beyond theoretical discussions about AI governance and into financial infrastructure.

If AI agents eventually direct trillions of dollars in commerce, knowing the customer will remain essential. Knowing the agent acting for that customer, and exactly what it was allowed to do, could become just as important.