Visa, Mastercard and Ant International build trust framework for AI agent payments

Visa, Mastercard, and Ant International are collaborating to create a common platform that helps identify and authenticate AI agents purchasing on behalf of their owners. This raises the broader question regarding autonomous software: if anybody would be able to demonstrate who gave the instructions to the agent, what was permissible for the agent to do, and who is liable for the faulty actions of the agent?

This is particularly relevant for cryptocurrency. Autonomous agents could potentially emerge as a new category of blockchain user since stablecoin payments, as well as programmable, always-on networks, function excellently with this kind of software.

On the other hand, on-chain transactions are more difficult to cancel as compared to card payments, making the consequences more serious in case something goes wrong and the agent makes a mistake or is hacked.

Three payment networks agree to recognize each other’s agents

As reported by Reuters, Visa, Mastercard and Ant International are in the process of working on a Know-Your-Agent interoperability framework that is aimed at assisting card networks, digital wallets and agent platforms to verify trusted agents across various systems without compromising their own risk controls.

The framework is expected to use Visa’s Trusted Agent Protocol, Mastercard’s Verifiable Intent and Ant International’s Agentic Mobile Protocol as points of reference. The company representatives claim shared standards can lead to reduced integration expenses, alongside enabling clients to manage risks in a more effective manner. The framework is being developed through BuildFin.ai, a platform launched by the Monetary Authority of Singapore.

The significance of this development is much wider than card payments since if agents start moving from one merchant, wallet, and blockchain to another, their own identity will have to be passed along with them.

Trust, not plumbing, is the sticking point

According to a joint report from Visa and Artemis, payments rails are only a partial solution to the problem. Traditional commerce assumes that there is always a person who is responsible for the final sale. Today, with the appearance of autonomous agents, that definition is blurred.

If a hacker redirects the agent’s spending, responsibility may lie with the user, the platform, the model provider, or the merchant. Chargeback rules were created for human-speed transactions and not for thousands of machine-to-machine transactions taking place among agents.

The US National Institute of Standards and Technology (NIST) is also studying the same issue. In a concept paper published in February, NIST suggested that there should be better controls for software agents concerning their identification, authorization, auditing, and accountability, including defense against prompt-injection attacks.

Why the money could end up on-chain

It is known that machines are able to deal at a high scale. According to Chainalysis, x402 payments on Base passed 100 million in three quarters comparing to almost no transactions in mid-2025. However, the company warned that most of the transactions were due to memecoin farming instead of sustained commercial demand.

The market size is already considerable. According to TRM Labs, the global retail crypto activity is approximately $979 billion for Q1 2026, giving agentic payments an already significant digital asset economy to integrate with as technology becomes more advanced.

Similar concepts are being investigated by central banks as well. In a working paper published by the Bank for International Settlements, it was demonstrated that the use of a general-purpose AI technology is able to perform some high-level intra-day liquidity management operations in a hypothetical environment of wholesale payment operations. However, the authors stressed the importance of reliability, accountability, human supervision and resilience against cyber threats.

The safety push is arriving with the capability

At the same time, companies involved in the development of frontier models are seeking stricter regulations. OpenAI has advocated for legislative regulation of safety involving AI technology based on its capabilities, which includes independent evaluation, cybersecurity measures and reporting of incidents.

According to reports by Reuters, calls for stringent regulations followed the cases in which sophisticated agents accessed external systems in ways that were not anticipated.

Meanwhile, the capability of AI models is evolving rapidly. OpenAI’s GPT-6 Astra safety report states that Astra is the first of its kind to reach the company’s Critical cybersecurity limit, meaning that it can find flaws that were previously unknown and create cyber attacks with minimal human supervision.

According to OpenAI, Astra is more resilient against prompt injection attacks and less likely to take destructive actions than GPT-5.6 Sol. However, its thinking process is more difficult to track and adversarial testing proved that it can sometimes avoid internal monitors.

Consumers continue to be careful. In a survey conducted by Cryptopolitan, 30.22% of respondents would not let their wallets be managed by an AI, though 25.9% would allow this from a reliable company.

According to Stanford University’s 2026 Artificial Intelligence Index, only 31% of people in the US believe that their government is capable of regulating AI in a responsible way, compared to the global average of 54%.

The major question for the future of cryptocurrencies is whether people will have faith in AI agents when it comes to handling money. Technology is advancing in this area, but the level of acceptance will depend on the establishment of an unambiguous regulatory basis which will determine who the agent acts for, what it is allowed to do, and who is responsible in the event something goes wrong.

The smartest crypto minds already read our newsletter. Want in? Join them.