The Know-Your-Agent framework aims to verify autonomous buyers across card networks, wallets and online marketplaces.
MARKET INSIDER — Visa, Mastercard and Ant International are developing a common identity and trust framework for artificial-intelligence agents capable of making purchases for consumers.
The proposed Know-Your-Agent system would allow card networks, digital wallets, AI platforms and merchants to recognize trusted agents across otherwise separate payment ecosystems. Each participant would retain its own approval, fraud-control and risk-management processes.
The initiative addresses a critical barrier to agentic commerce: before allowing software to complete a transaction, merchants and payment providers must establish which agent is acting, who authorized it and whether the purchase remains within the customer’s instructions.
Key Highlights
- The framework would create shared principles for identifying and onboarding AI agents across cards, wallets, marketplaces and agent platforms.
- It combines elements of Visa’s Trusted Agent Protocol, Mastercard’s Verifiable Intent and Ant International’s Agentic Mobile Protocol.
- The collaboration could lower integration costs, but it does not yet establish universal rules for liability, disputes or unauthorized purchases.
From shopping recommendations to autonomous purchases
Consumer AI systems already help users research products, compare prices and recommend travel or restaurant options.
The next stage is allowing an agent to complete the transaction.
A customer could instruct an AI assistant to book the lowest-priced direct flight within a defined schedule, reserve an approved hotel and pay with a designated card. The agent would then search multiple platforms, evaluate options and execute the purchase without requiring the user to complete every checkout step.
That convenience introduces a new trust problem.
Traditional online commerce assumes that a person is interacting with a merchant’s website or application. Fraud systems distinguish genuine customers from automated bots partly because bots are frequently associated with credential theft, inventory manipulation or payment fraud.
Agentic commerce requires merchants to accept some automated buyers while continuing to block malicious ones.
The Know-Your-Agent framework is intended to provide the identity layer needed to make that distinction.
What Know Your Agent must establish
Know Your Agent, or KYA, is related to—but different from—the Know Your Customer procedures used by banks and other financial institutions.
KYC verifies the identity and risk profile of a person or legal entity. KYA must connect several additional elements:
- the identity of the AI agent and its operator;
- the customer who authorized it;
- the specific powers delegated to the agent;
- the limits applying to the transaction; and
- a verifiable record of what the customer intended.
A valid agent may, for example, have permission to purchase airline tickets costing less than $1,500 but not to book first class, use another payment card or modify the trip dates.
The payment system must prove that the transaction remained within those boundaries.
The proposed framework would help networks recognize agents approved through another participating ecosystem. A wallet would not necessarily have to repeat the entire onboarding process for an agent already carrying trustworthy credentials from a card network or marketplace.
However, interoperability does not mean that every participant must accept the same risk. Banks, wallets and merchants would continue applying their own transaction controls and deciding whether to authorize individual purchases.
Three protocols form the technical foundation
The collaboration builds upon systems already developed independently by the three companies.
Visa’s Trusted Agent Protocol uses cryptographic signatures to help an AI agent prove its identity and authorization directly to a merchant. Signatures can include timestamps, session identifiers and information binding the interaction to a specific merchant and webpage, helping prevent impersonation and replay attacks.
Mastercard’s Verifiable Intent creates a tamper-resistant record connecting the customer, the delegated agent and the approved transaction. The system is intended to prove not only who the agent is, but whether its action matches the consumer’s original instructions.
Mastercard describes this as a cryptographic chain covering consumer identity, delegated authority and checkout integrity. The company has developed Verifiable Intent with Google and opened the specification to broader industry contributions.
Ant International’s open-source Agentic Mobile Protocol is designed around mobile wallets, super apps and cross-platform commerce. Its KYA component establishes an agent’s digital identity and certifies what the agent is authorized to do.
Combining elements of the three systems could reduce the need for merchants and AI developers to create separate integrations for every card network and wallet.
Why interoperability matters
The payments industry has seen this problem before.
Online card payments expanded more rapidly once merchants, banks and networks adopted common technical and security standards. Tokenization, contactless payments and digital wallets also became easier to scale after systems learned to recognize credentials across different platforms.
Without interoperability, an AI agent approved for a Visa transaction might require separate registration and verification before using Mastercard, an Ant-linked wallet or an online marketplace.
That fragmentation would raise development costs and favor the largest AI platforms, which can afford dozens of customized integrations.
A common KYA layer could allow smaller developers to build agents that work across multiple payment systems. Merchants could also receive standardized identity and authorization signals without surrendering control over their own fraud models.
The companies say this should improve risk visibility and shorten the time required to launch agentic services.
The framework remains an industry initiative rather than a final, universally adopted standard. Technical specifications, governance arrangements and implementation timelines have not been fully disclosed.
Fraud and liability remain unresolved
Identifying a trusted agent does not answer every legal and commercial question.
An agent may be genuine but still make an incorrect purchase. It could misunderstand a customer’s instructions, rely on inaccurate product data or select an option influenced by undisclosed commercial incentives.
Disputes will become more complicated when several parties are involved: the consumer, AI developer, agent platform, merchant, wallet, issuing bank and payment network.
If an agent books the wrong hotel, responsibility could depend on whether the customer’s mandate was ambiguous, the AI reasoned incorrectly or the merchant provided misleading information.
Existing chargeback systems were designed largely for unauthorized payments, undelivered goods and merchant disputes—not software acting with partial but imperfect authority.
KYA provides evidence that could help allocate responsibility. A cryptographic record can show which agent acted, what limits were established and whether the completed purchase matched the approved checkout.
But the framework does not by itself determine who absorbs the loss.
Industry adoption will therefore depend on clear rules covering consumer consent, error correction, refunds, data privacy and liability for agent mistakes.
Singapore emerges as an agentic-finance hub
Visa, Mastercard and Ant International will conduct the collaboration through BuildFin.ai, an initiative convened by the Monetary Authority of Singapore.
BuildFin.ai brings together financial institutions, technology companies and researchers to develop AI applications for financial services.
The platform has already supported work on Safeguards for Agentic Finance at Runtime, or SAFR. That approach examines how an AI agent’s proposed action can be authorized, monitored and recorded before execution.
Its potential applications include payments, treasury operations, trading, credit decisions and insurance claims.
Singapore is a logical venue because it combines a sophisticated financial center, strong regulatory institutions and extensive connections to Asian digital-wallet ecosystems.
Ant International also brings experience from markets where mobile wallets, super apps and QR payments are more deeply integrated into everyday commerce than in many Western economies.
That makes the initiative particularly relevant for Southeast Asia, where agentic transactions may need to move between cards, domestic payment systems and multiple cross-border wallets.
Strategic implications for payment companies
Agentic commerce presents both an opportunity and a threat to established payment networks.
If AI agents increase the number and frequency of digital purchases, Visa and Mastercard could process additional transactions and earn more network fees.
But AI platforms may also gain control of the customer relationship. Consumers could interact primarily with an assistant rather than a bank, card brand or merchant.
By establishing the identity and authorization standards, payment networks can preserve their role as trusted infrastructure even if the visible shopping interface shifts to AI.
Ant International benefits by ensuring that Asian wallet networks participate in the emerging standard rather than operating as extensions of card-based systems designed elsewhere.
The initiative also creates strategic value from transaction data. Risk signals concerning an agent’s identity, reputation and past behavior could become essential to fraud prevention.
Competition will focus not only on which company offers the best consumer AI, but on which institutions provide the trust credentials that merchants are willing to accept.
What investors should watch
The first test will be whether major AI developers, marketplaces, banks and merchant processors adopt the framework.
Participation from companies such as Google, Microsoft, OpenAI, Amazon, Shopify and leading wallet providers would increase the likelihood that KYA becomes an industry standard rather than another competing protocol.
Investors should also watch for real-world transaction pilots and performance data covering fraud rates, approval rates and disputed purchases.
Regulatory treatment will be equally important. Authorities must decide how existing consumer-protection, payment-services and data-privacy rules apply when autonomous software makes a purchase.
Finally, liability rules will determine the economics. If payment networks or issuers must absorb a disproportionate share of agent-generated errors, the costs could outweigh the additional transaction volume.
The collaboration between Visa, Mastercard and Ant International is an early attempt to build the identity infrastructure for a market that does not yet exist at scale.
Agentic commerce will not expand simply because AI can shop. It will expand when consumers, merchants and banks can verify that an autonomous buyer is legitimate, authorized and accountable.
