AI agents could soon shop and pay for you. Visa, Mastercard and Ant want to make them trustworthy

Artificial intelligence agents are moving beyond answering questions and recommending products. They are increasingly being designed to search, compare and eventually buy products on behalf of consumers, bringing a new challenge to the payments industry: how do banks, card networks and merchants know that an AI agent is authorised to spend a user’s money?

Visa, Mastercard and Ant International are now working on a common framework to address that problem, as the payments industry prepares for the growth of so-called agentic commerce.

The companies said on Thursday they had begun collaborating on a Know-Your-Agent interoperability framework that would allow card networks, digital wallets, AI agent platforms and online marketplaces to recognise trusted AI agents across different payment ecosystems.

The initiative is aimed at allowing each participant to maintain its own approval and risk-management systems while creating a common way to establish the identity and trustworthiness of an AI agent.

The development comes as AI moves closer to becoming an active participant in transactions rather than simply a tool used by consumers to find information.

When AI becomes the buyer

In conventional online shopping, a consumer searches for a product, selects it, enters payment details and authorises the transaction.

More from Business

An AI agent could eventually perform much of that process itself.

A user could instruct an agent to find a particular product below a certain price, compare alternatives and complete the purchase. The agent would then act within the permissions given by the user rather than simply providing a recommendation.

That changes the risk model for payments.

A payment provider would need to establish whether the agent is genuine, who authorised it, what it is permitted to purchase and whether the transaction falls within those permissions.

The problem becomes more complicated when an AI agent, merchant, wallet and payment network operate on different systems.

The framework being developed by Visa, Mastercard and Ant is intended to help those systems recognise trusted agents without requiring every participant to create separate connections with every other participant.

The companies said the initiative could reduce integration costs and provide greater visibility into risks, potentially accelerating the development of new agentic payment services.

The framework builds on existing technologies developed separately by the three companies, including Visa’s Trusted Agent Protocol, Mastercard’s Verifiable Intent and Ant International’s Agentic Mobile Protocol.

Why trust is becoming a payments problem

The International Monetary Fund has warned that agentic AI could require changes to the way payment systems handle authorisation, compliance and risk.

In an April 2026 note, the IMF examined how autonomous AI systems could affect payments, including authorisation, liquidity, settlement, compliance and resilience. It highlighted a central problem: AI systems can make probabilistic decisions, while payment infrastructure generally depends on deterministic rules, clear authorisation and predictable settlement.

That creates a distinction between what a consumer intends an AI agent to do and what the agent actually does.

For example, a user may authorise an agent to purchase a product up to a particular value or only from certain merchants. Payment systems would need mechanisms to establish whether a transaction remained within that authority.

That is why identity, permissions and spending limits are likely to become increasingly important alongside traditional payment authentication.

The issue is not only whether an AI agent can make a payment. It is whether other participants in the transaction can verify which agent acted, who authorised it and what the agent was allowed to do.

India is preparing for the same shift through UPI

The move towards agentic payments is not limited to global card networks.

India is also preparing a framework that could allow AI agents to conduct small-value UPI transactions without requiring users to approve every individual payment, Reuters reported last week, citing people familiar with the matter.

The proposed Unified Agent Protocol is expected to initially focus on low-value and routine purchases. It could include rule-based payments, spending limits, identity checks and provisions dealing with liability.

The proposed framework could also use mechanisms such as UPI Circle and Reserve Pay to allow users to delegate funds or block money for specific purposes, Reuters reported.

The scale of India’s payments network makes the development significant. UPI processed 24.51 billion transactions worth Rs 29.82 trillion ($314.21 billion) in August 2026, according to the Reuters report.

That means India’s experiment with agentic payments could provide a large real-world environment for testing how delegated authority, transaction limits and AI-led payments work.

The development also comes as global payment companies explore similar systems in India, Reuters reported.