Banks navigate AI risks solo as regulators play catch-up

• Key insight: AI agents are creating a new gray area for banks around fraud and liability, leaving financial institutions to their own devices in determining best how to protect customers and themselves in the absence of clear rules. 
• Expert quote: “Statutory changes are hard and take a long time, and nobody wants to wait.The banks don’t want to wait because they like the idea of there being more payments and less friction in payments. Customers don’t want to wait because they love AI doing this for them. Merchants don’t want to wait because they like selling things.” — Julie Hill, Dean, University of Wyoming College of Law.

WASHINGTON — The rapid emergence of AI agents and their growing adoption among consumers have created a new set of risks for banks to contend with — and they’re largely on their own to mitigate those risks.

Processing Content

Without established rules of the road for AI agents, banks are left to navigate the technology’s promises and pitfalls in real time while preserving the safety and soundness of their systems.

Industry insiders expect that prudential regulators will eventually address these questions around AI agents. Until then, banks will have to develop their own protocols to mitigate risks and safeguard their systems.

“The agents are here now, and rulemaking takes a while,” said David Zaring, a professor of legal studies and business ethics at the University of Pennsylvania’s Wharton School of Business. “In the medium-future, or in the short-term future, I think banks may be walking the tightrope without much of a net.”

AI agents are bots that have been empowered by a consumer to make financial decisions on their behalf — actions ranging from tracking down hard-to-find concert tickets to completing online purchases. While the potential benefits for consumers and commerce are potentially significant, the technology also introduces a range of risks for banks.

Those risks present themselves in two distinct flavors: fraud on the one hand and faster deposit outflows on the other. While industry stakeholders see both as concerns, many view fraud as the more immediate threat as banks grapple with determining whether transactions initiated by AI agents are legitimate 

Justin Herring, a partner at Mayer Brown, said that, from a technical standpoint, it can be nearly “impossible” to determine whether a request is coming from a bad actor or an actual customer.

“It’s one thing if a threat actor is hacking directly into a bank, but when the request is coming from a customer, it may be technically impossible to detect,” he said. “The request might actually be from a real customer, or not. You don’t really know who it’s from.”

In response, some of the biggest U.S. banks, including JPMorgan Chase, Citi, Bank of America,Wells Fargo and Morgan Stanley, are blocking external AI agents from accessing their customers’ accounts.  

The restrictions reflect a broader effort by banks to establish where liability falls when AI agents initiate unauthorized or disputed transactions. JPMorganChase, for its part, has said customers who authorize AI agents to act on their behalf will need to understand the potential liability that comes with delegating that authority.

Julie Hill, dean of the University of Wyoming College of Law, said existing rules generally favor consumers, who typically are not held liable for unauthorized transactions. But the growing use of AI agents complicates that framework.

“There’s the bad actor problem, but another problem with AI agents is, say I ask it to buy me a pair of shoes and something goes wrong, so I want to dispute my bank charge because I didn’t authorize it, and my bank says, ‘But you did authorize it because you gave the AI agent authority to do this,'” Hill said. “I think that the concern there is that if you can’t show that the transaction was authorized, the liability for the bank goes way up, and that’s a problem in the AI context.”

Read more:

In the absence of regulatory guidance, banks will have to establish their own guardrails, industry stakeholders say. That could include requiring customers to disclose when they are using an AI agent and confirm that they have authorized specific transactions.

“I think the first effort here is going to be made by the banks, and they’re going to try to figure out a way to have customers say, ‘I authorize AI agent X to act on my behalf,’ and how the bank confirms that, and how the AI agent shows that it was authorized,” Hill said.

Future regulations could help resolve some of the uncertainty by establishing standards for determining whether transactions initiated by AI agents were properly authorized and who is liable when something goes wrong.

“You can imagine a set of rules that would provide some certainty and some consistency,” Herring said. “We already have that from a non-AI context coming from banking regulators, from the CFPB, from statutes at both the state and federal level. It’s just that none of them address this current situation. So I think the rules of the road will help. They usually lag industry practices a bit.”

The emergence of agentic AI has caught the attention of prudential regulators. In late October, Comptroller of the Currency Jonathan Gould expressed confidence in banks’ ability to defend against malicious AI activity.

“Banks are very focused on defending the frontier,” he said during an appearance on Fox Business’ Varney & Co. “Banks are some of the most fortified parts of our system, and we have been, across the U.S. government, very focused on making sure that they have the latest tools so they can defend themselves.”

While regulators are working to assess the risks posed by AI agents, the technology’s adoption is unlikely to wait for a formal regulatory framework. Banks, consumers and merchants all have incentives to embrace the technology, Hill said. 

“Statutory changes are hard and take a long time, and nobody wants to wait,” she said. “The banks don’t want to wait because they like the idea of there being more payments and less friction in payments. Customers don’t want to wait because they love AI doing this for them. Merchants don’t want to wait because they like selling things.”