Payment networks are moving aggressively to establish the technical standards and authorization protocols for autonomous “agentic commerce,” seeking to secure their position at the center of AI-driven retail before clear legal liability rules and broad consumer acceptance take shape.
Mastercard Inc. announced a payment system designed to integrate directly with artificial intelligence agents. Consumers can issue specialized virtual card numbers to AI assistants, enabling the software to select and purchase items online without requiring human confirmation for every transaction.
The system allows users to set parameters such as maximum spending limits, approved merchants, and optional pre-checkout approval requirements. Meta Platforms Inc. developed “Muse,” an AI tool designed to search for products and navigate online checkout pages, but Meta’s system requires the user to give final explicit approval before a payment is processed.
Visa Inc. has been pursuing a parallel strategy. The company partnered with Web3 and infrastructure provider Alchemy earlier this year to support automated transactions and has begun deploying its own platform, Visa Intelligent Commerce.
The drive to build infrastructure for software bots capable of shopping independently comes as card networks look to prevent technology companies from establishing alternative payment rails. The rush to define agentic commerce standards mirrors past shifts in electronic payments.
Over recent decades, payment networks retained control of online and mobile shopping by deploying network tokenization, replacing 16-digit primary account numbers with dynamic digital tokens. They also established global transaction standards through EMVCo, the body jointly owned by major card brands including Visa, Mastercard, American Express, and Discover.
“This is a land grab for infrastructure standards,” Phil Bruno, chief strategy and growth officer at payment systems firm ACI Worldwide, said. “If they set the standards for agentic commerce, they can keep the commerce in their environments for decades to come.”
To address potential transaction disputes arising from automated purchases, Mastercard developed “Verifiable Intent,” a digital record system that logs the user’s explicit instructions, spend limits, and permissions prior to an order being placed.
Despite the infrastructure rollout, open questions remain regarding legal responsibility when autonomous software makes purchasing errors. It remains legally unclear how federal regulators and card network dispute mechanisms will treat purchases resulting from software hallucinations, system glitches, or unexpected AI decisions.
In the United States, consumer financial protection is governed by federal statutes written for human-directed transactions. Under the Truth in Lending Act and its implementing Regulation Z, consumer liability for unauthorized credit card charges is generally capped at $50, and major card networks typically enforce voluntary “Zero Liability” policies. Debit card transactions are regulated under the Electronic Fund Transfer Act and Regulation E.
When asked who bears financial responsibility if an AI agent makes an incorrect, fraudulent, or unauthorized transaction, Mastercard pointed to its Verifiable Intent digital audit trail as the primary tool for resolving claims. The company did not specify how liability would be divided among card-issuing banks, merchants, software developers, or cardholders if an agent buys an item outside prescribed user parameters.
Mastercard stated that transactions processed through its agentic payment systems will maintain the standard protections associated with traditional card payments. Mastercard characterized its investment as preparation for long-term shifts in retail behavior, framing the integration of artificial intelligence into commerce as an inevitable evolution.
Market research indicates that consumer willingness to delegate purchasing power to software remains low. A study commissioned by ACI Worldwide evaluating fashion shoppers in the United States and the United Kingdom found that only 7% of respondents were willing to allow an AI assistant to complete purchases without explicit approval under predefined rules.
More than 50% of surveyed consumers stated they were uncomfortable allowing AI tools to make financial transactions on their behalf. Dan Coates, product management director at ACI Worldwide, attributed the hesitancy to a desire for direct control at the point of sale. “We want to be the ones to make that final click,” Coates said.
The survey found that consumers prefer AI tools that assist with research rather than execution. Thirty-five percent of respondents identified price-drop alerts as a valuable feature, and 35% cited cross-retailer price comparisons. Personalized product recommendations were cited by 18% of respondents.
“AI is a transformational technology,” Mastercard said in a statement. “It’s a matter of when, not if, people use it.”
