Even as financial institutions begin integrating AI agents into existing payments rails, it might be a good idea to develop industry standards that allow the agents to talk to one another, industry experts say. 

Agentic AI payments tools are AI agents which can plan and execute a payment on behalf of the customer after receiving specific instructions. Mastercard and Visa are among some companies that are planning on deploying the tech globally this year, according to FinAi News’ prior reporting.

The industry should not build agentic on top of the existing e-commerce framework, John Lunn, chief executive of payments provider Gr4vy, told FinAi News. 

“Payments rails have been built on the ‘beware of bots’” concept, but as AI agents proliferate, there is now the idea that “some bots are good,” he said. On the contrary, “you are essentially launching a heavy nuclear bomb on top of something that’s just not ready to handle it,” Lunn said. 

Potential solutions

Other tech experts disagree.  

Integrating agentic AI payments into existing payments rails is not only possible, but easier than creating a separate channel for AI agents, Sovan Shatpathy, senior vice president of product management and development at Oracle Financial Services, told FinAi News. If agents have a common method of communication, they can act without disrupting the system, he said. 

“If you’re invoking an agent from another company or another bank, you just need to have some basic common protocols so that those guys can do a handshake,” Shatpathy said. Oracle plans to launch 130 AI agents by May.  

Meanwhile, Greg Hodges, head of trust and safety at JPMorgan Payments, told FinAi News that the $3.6 trillion bank is evaluating multiple methods to tackle fraud and safety concerns around agentic payments. 

Most data elements — such as device identification or geographic location — used by financial institutions to determine whether payments transactions are fraudulent or not are irrelevant or not even available in an agentic commerce world, Hodges said. 

Rails, languages vary 

Gr4vy’s Lunn said merchants and FIs that aim to deploy agentic AI must build different gateways for agentic payments to be able to segregate them away from traditional methods. 

“Businesses are concerned about fraud that comes with agentic payments,” he said. “By creating different rails, we can develop anti-fraud methods designed specifically for agents and drive adoption.” 

One alternative being studied is for each agent to have its own wallet and make transactions from that specific wallet to avoid overspending and eliminating the need for the agent to use a customer’s credit card, Lunn said. 

Agentic payment providers can also help businesses create rules and benchmarks that each business can accept a certain number of payments from each agent, under a certain dollar amount, to reduce bot spams and chargebacks, Lunn said. 

Over time, the financial services industry will develop industry standards on how agents communicate and transact, Hodges said, adding that “some certainty” is needed before deploying agentic AI payments. 

“There’s so much uncharted territory here that we have to resolve,” Hodges said. “It will be interesting to see is, you know how that process works out, what’s agreed upon and what’s not.” 

Register here for the inaugural FinAi Banking Summit, taking place March 2-3 in Denver. View the full event agenda here.