As real-time payments go global, how can banks manage AI security?

Key insight: Experts from the Federal Reserve, JPMorganChase and Flywire say new forms of AI are adding risk to cross-border payments.  

What's at stake: Banks face competition from fintechs to capture international payments share. 

Expert quote: "Cross border payments do tend to be complicated." —Flywire's Mohit Kansal

As the Federal Reserve inches toward taking its FedNow transfer rail global, it's also considering how the increasing use of artificial intelligence in financial services is changing payment processing. 

"Our focus has been on growing domestic reach. We're super excited to be entering into the space of cross border," Bernadette Ksepka, senior vice president and deputy head of product management for the Federal Reserve System's FedNow payment rail, said at a recent Boston Fintech Week event.

FedNow, which recently passed the 1,700 membership mark, is considering a change that would allow FedNow to support cross-border transactions by using intermediaries that are not Federal Reserve banks to complete real-time cross-border transactions."We're taking a crawl-walk-run approach," Ksepka said. 

Innovation on two fronts

Expanding real-time payments to cover cross-border transactions may also boost security risk. The expansion comes as banks are also embracing agentic commerce and other new forms of AI. In Boston, Ksepka spoke on a panel that considered the risk and opportunities of agentic AI. 

The panelists said vetting international payments as AI agents take on a larger role in banking will become more difficult but necessary given the opportunities to gain share of a fast-growing market. Visa and Mastercard, for example, have both entered into partnerships designed to expand their shares of international payments as both card brands also invest heavily in agentic commerce. The parallel trends will add to risk-management challenges, according to Boston-based payment fintech Flywire.

"Cross border payments do tend to be complicated," Mohit Kansal, Flywire's chief payments officer, said at the Boston fintech event. 

Kansal said each country has its own set of regulations, and those rules keep changing. 

AI can help speed that work, but only up to a point, according to Kansal. "I wouldn't let AI update our system without having an expert validate it," Kansal said.

Flywire specializes in processing cross-border tuition payments, though it has diversified in recent years to add health care and other industries. These transactions tend to be larger and will likely not be executed autonomously by AI agents, according to Kansal. Agentic AI will instead make the work that surrounds the payment easier.

"Everybody wants to experiment with AI and AI agents," Kansal said. "[But] I don't see a lot of agentic payments going through. The AI agents will go in and simplify the workflow first."

Agentic commerce can also help banks compete with nonbank rivals that have made inroads into international payments in recent years.

"Fintechs have completely disrupted cross-border payments and obliterated checks," Jennifer Perry, co-head of innovation technology at JPMorganChase, said during the Boston Fintech event. "Our clients want to figure out how to make and receive payments faster, but they're also really thoughtful about how they reconcile. The biggest component is trust."

Good versus bad

New forms of AI will create risks for banks, including AI agents' ability to evade security controls.

"To see that growth in autonomous payments, you need safe, secure, always on, data rich infrastructure to support that," Ksepka said. "Banks need to start by thinking about the infrastructure."

Writing for American Banker, Pat Kinsel, CEO and founder of identity verification firm Proof, said government efforts to guide agentic AI security are running behind the pace of the risk.

"Our current regulatory frameworks need to catch up. Frameworks like 'know your customer,' or KYC, and Suspicious Activity Reports, or SARs, were all established under the assumption that humans were the ones initiating transactions and filling out forms. Not only does that assumption break down in an AI-mediated world, but much of our existing regulatory infrastructure was already failing to respond to modern fraud threats," Kinsel said.

Panelists at the Boston fintech event said vetting data and users are vital to managing security for agentic commerce. 

"Anything in AI is only as good as the data," Kansal said. "If you don't have the right data, that's when you get an exception and a human goes in and tries to get the data from [more than a dozen] different sources."

Regarding AI agent identity, Visa and Mastercard recently launched designs to support "know your agent" strategies at banks. The card network products support agentic shopping, make products discoverable across AI ecosystems and support "trusted" agentic transactions. 

"AI is going to make payments more autonomous." Ksepka said "Fraudsters are also using AI to fuel more sophisticated fraud. We need to make sure the good AI is prevailing over the bad AI."

—Chana Schoenberger and Megan Ryan contributed to this story.