Visa, Mastercard and Ant will share AI-agent identity checks. They still won't say who pays when an agent is wrong.

On September 9, 2026, Ant International, Mastercard and Visa announced they had begun collaborating on a Know-Your-Agent (KYA) interoperability framework. Reuters reported the initiative the next day. The goal is shared standards to identify and verify AI agents that make purchases on users' behalf, so card networks, digital wallets, agent platforms and marketplaces can recognize trusted agents across payment ecosystems.

This is a planning announcement, not a product launch. The press release's own language says the three companies "will now explore opportunities to work towards common principles." There is no specification, no published timeline, and no participating agent platform or marketplace named yet.

What the companies say the framework will do

The announcement centers on three pillars. Cross-network operator traceability: each agent linked to a validated operator, cardholder or business, so activity is attributable. Shared certification requirements: agents assessed against security and behavioral requirements. Continuous transaction monitoring: ongoing evaluation using identity and transaction signals. All three are vendor claims from the joint release, not independently verified designs.

The work builds on three existing protocols. Visa's Trusted Agent Protocol uses cryptographic signatures so an agent can prove identity and authorization to a merchant, with timestamps and session identifiers to resist replay. Mastercard's Verifiable Intent creates a tamper-resistant record connecting the customer, the delegated agent and the approved transaction. Ant International's open-source Agentic Mobile Protocol covers wallets and super apps, with its own KYA component certifying what an agent may do. Each network keeps its own approval and risk-management processes.

The collaboration will run through BuildFin.ai, an industry platform convened by the Monetary Authority of Singapore that brings together financial institutions, technology providers and researchers to develop AI for financial services. The companies say the work builds on Singapore's SAFR framework, Safeguards for Agentic Finance at Runtime.

Why it matters for builders

The commercial argument is onboarding cost. Without interoperability, an agent approved for Visa transactions would need separate registration and verification for Mastercard, an Ant-linked wallet, or a marketplace. The press release's framing: shared trust signals reduce duplicate integration work and speed up time to market for agentic services. Market Insider's analysis adds the other side of the same coin: fragmentation would favor the largest AI platforms, which can afford dozens of custom integrations, so a shared layer would help smaller developers build agents that work across payment systems.

There is also a power question underneath. Phil Bruno of ACI Worldwide told Fortune this is "a land grab for infrastructure standards." Whoever's verification becomes the default also shapes who controls the checkout relationship as agent-initiated transactions scale. Builders should read the announcement as positioning, not charity.

What it does not solve: the liability gap

The central unanswered question is what happens when an authorized agent authorizes the wrong transaction. A genuine agent can still misread instructions, rely on bad product data, or act on undisclosed incentives. Existing chargeback systems were designed for unauthorized payments and undelivered goods, not software acting with partial but imperfect authority.

The companies have not answered it. When Fortune asked Mastercard who would be responsible if an agent made an incorrect, fraudulent or unauthorized purchase, Mastercard pointed back to the Verifiable Intent record. It did not name a party accountable for losses. A cryptographic record shows which agent acted and what the mandate was. It does not decide who absorbs the loss.

Consumer behavior adds a second drag. Research commissioned by ACI Worldwide and conducted by YouGov found that just 7% of surveyed U.S. and U.K. fashion shoppers would let an AI assistant buy without their approval, and more than half were uncomfortable letting AI purchase on their behalf at all. The infrastructure is being built for a user who is not yet willing.

What a founder should do now

Do not build against KYA yet. There is nothing to integrate with. The practical moves are cheaper: track the three constituent protocols (Verifiable Intent's spec is open for industry contributions), design your agent's purchase flow to produce a consent record that links customer, agent and mandate, and assume that every participant will still run its own risk checks.

Note the adjacent product moves, which are real while KYA is a plan. Mastercard launched a consumer AgentCard virtual card on September 17 through Alchemy, letting cardholders link an AI agent to a virtual card with spending caps, retailer restrictions and optional pre-checkout approval. That is a product a founder can study today. KYA is a standards conversation to watch.

What remains unknown

No technical specification, governance arrangement or implementation timeline has been disclosed. No agent platform, marketplace or merchant processor has been named as a participant; without adoption by companies like Google, Amazon or Shopify, KYA stays an industry initiative rather than a standard.

Liability and loss allocation are unresolved, and the companies have declined to name who is accountable. Revenue models are unstated. Regulatory treatment is open: authorities still have to decide how consumer-protection and payment-services rules apply when autonomous software completes a purchase. And the McKinsey projection cited by the companies, that agents could orchestrate $3 trillion to $5 trillion in consumer commerce by 2030, is a forecast, not evidence of demand.

The identity layer for agentic commerce is now officially on the industry's agenda. The money layer, who pays when it breaks, is not.