- Key insight: Mastercard and Visa have updated agentic commerce tools to bolster AI agent risk.
- What's at stake: The card brands are trying to address concerns about AI agents going outside of their instructions.
- Expert quote: "Agentic commerce will only scale at the speed of trust," —Sherri Haymond, Mastercard.
With buzz over
"There are definitely things that have to be addressed. There is potential risk if agentic AI is left unchecked," Sherri Haymond, executive vice president of digital commercialization at Mastercard, told American Banker. Mastercard has updated its suite of AI tools to address trust and risk, while Visa has made similar moves. At the same time, Visa, Mastercard and Ant International are teaming on a protocol to vet AI agents. The card networks aren't alone, as
"Agentic commerce will only scale at the speed of trust," Haymond said.
No secret agents
As part of its
"If you are interacting with someone or an agent that is acting on someone's behalf, there needs to be transparency about who owns that agent and where that agent came from," Haymond said.
Agent Connect offers an integration point that helps AI agents reach merchants, commerce services and payment providers. Merchants choose which AI experiences to connect with while maintaining control over pricing, fulfillment, customer relationships, brand experiences and business rules, with Mastercard providing governance, monitoring and controls.
AI agents discover merchant-provided product catalog information such as product details and availability, then recommend products for a shopper's cart and confirm final pricing, taxes shipping costs and fulfillment details.
"Agent Connect is a very important step for us," Haymond said. "We have been working on filling those gaps when you aren't there in person."
The transaction is then completed by using secure payment credentials from any network. A related product, Agent Pay, uses Verifiable Intent to help ensure purchases are completed securely with the consumer's authorization. Verifiable Intent, developed by Mastercard and Google, binds an AI agent's actions to a user's stated intent, creating a chain showing what the user authorized and what the agent is allowed to execute.
Mastercard's primary rival, Visa, in early September released research that found only 23% of U.S. consumers trust generative AI to handle payments on their behalf. Consumers have more trust in legacy payment brands.
"Every major shift in commerce has been built on a foundation of trust, and agentic commerce will be no different," Oliver Jenkyn, Visa group president, told American Banker. "As AI moves from recommending what to buy to transacting for consumers, the brands people trust will matter more than ever."
Visa's AI risk products include the Visa Intelligence Commerce platform. This includes agentic commerce tools Trusted Agent Protocol and Intelligent Commerce Connect. "Together, they directly address the trust gap," Jenkyn said.
Large payment companies are also partnering to address AI risk.
Ant International, Mastercard and Visa have begun collaboration on a know-your-agent, or KYA, interoperability framework designed to help card networks, digital wallet ecosystems, agent platforms and marketplaces streamline agent onboarding and identification across networks, based on shared principles while preserving each network's own verification and decisioning processes.
The Ant/Visa/Mastercard collaboration will focus on cross-network operator traceability, with each agent linked to a validated operator, cardholder, or business/organization, enabling clear attribution of agent activity. Each agent will also be assessed against security and behavioral requirements to ensure it operates as expected. There will additionally be continuous monitoring of AI agents with evaluation using a combination of identity and transaction-related signals to support ongoing assessments and certification.
AI agents and VAS
For the card networks, agentic commerce could be a source of greater payments volume, but also a way to generate value-added-service revenue as Visa and Mastercard use their networks to support risk management.
"As AI agents increasingly transact on behalf of consumers, we believe the value of identity, authentication, fraud prevention, and other value-added service capabilities should increase rather than diminish," BofA Research said in an analyst note on Visa and Mastercard. "The market continues to view stablecoins and agentic commerce as potential disintermediation risks, but recent developments suggest the networks are becoming key beneficiaries instead."
But vetting agent identity may not be enough.
"Knowing that the agent is legitimate is a good start, but you still need to know whether it's doing what the customer asked," Louis Hoch, CEO of payments company Usio, told American Banker. "An agent can be trusted and still misunderstand an instruction or spend more than it was allowed to. Before a payment goes through, there needs to be a check that the purchase falls within the permission the customer actually gave."
Read more:
OCC and FDIC finalize narrower bank supervision procedures As education costs rise, credit unions chip in for employees A $111 million scheme to keep bad merchants banked How Google and Apple's youth plays are pressuring banks
One option is to build a separate digital wallet for AI agents, according to Hoch.
An agent gets a set amount of money to work with, instead of access to the user's primary bank account or full credit line. "You should be able to see what it's spending and shut off its payment access immediately," Hoch said. "People are more likely to trust an agent when they know how much it can spend and that they're still in control."
The industry also needs clear answers for what happens when an agent makes a mistake, according to Hoch.
"Just because a payment went through doesn't mean the agent bought what the customer wanted," Hoch said. "There needs to be a record of what the customer asked for, what the agent was allowed to do and what it actually bought. That helps everyone work out whether it was fraud, a misunderstanding or something else, and who needs to put it right."
