The CEO of a fintech that works with big banks expects it’ll be about six months before lenders establish clear protocols around what agents can and can’t do on behalf of customers.  

As AI agent use takes off and banks consider access questions, “there’s a lot of code reds going on in the banks right now,” Socure CEO Johnny Ayers said in a recent interview. 

Socure, an identity and risk intelligence provider, works with about 3,000 banks, including 19 of the top 20 in the U.S., and about 700 fintechs, Ayers said.  

“I think we're still a ways out from the banks having a clear protocol around when [a person can] go open a new bank account or a credit card with her agent,” Ayers said. 

AI agents such as Meta’s Muse, Instinct and OpenAI’s dots, acting on behalf of consumers, have quickly proliferated in recent weeks, prompting banks to throw up walls and bank CEOs to warn of potential liability issues. 

JPMorgan Chase, Bank of America and Citi have all blocked AI agent access to customer accounts. Yet Axios has reported top U.S. banks have discovered AI agents within accounts anyway, with some agents taking actions that went beyond their instructions.

Banks are “built on standards and protocols, and, I think, are probably taking a defensive stance while they get the rules and protocols with the networks in place,” Ayers said.  

There’s plenty for banks and other industry players to sort through and establish parameters around, as well as how regulators might respond if there are customer issues, Ayers said. Meanwhile, the technology is evolving rapidly. 

“You're seeing the vulnerabilities in cyber. People are kind of scared,” Ayers said. “Folks are definitely taking more of a wait-and-see [approach].”

“You have thousands of people in these banks that are working to basically patch all of the critical vulnerabilities that [Anthropic’s Claude] Mythos and others are highlighting. And so that's not an organization they generally want agents running rampant in,” he added. 

The fraud, anti-money laundering and know-your-customer framework is not yet written for agents, he noted. “Can an agent open a bank account on your behalf? What is it allowed to do? What are its limits?” 

“There’s an acceleration right now” on the bank and card issuer side, “to try to figure out what agents are allowed to do,” Ayers said.

There are also data access concerns. 

“You have this rightful blocking to make sure that these agents aren't going in and basically siphoning off transaction data on behalf of consumers, where consumers are unknowingly giving access,” Ayers said.  

Banks have more control over mobile use, Ayers said, but with web applications, “they need to figure this out now, because I can let an agent take over my browser entirely.”

He pointed to a juxtaposition between fintech and bank approaches. The latter are “at such big scale, they have the luxury of waiting and seeing,” Ayers said, where fintechs such as Mercury are taking the opposite approach. 

Bank reactions

PNC CEO Bill Demchak, at the Money20/20 conference in Las Vegas last month, said agent-only transactions will never happen. During a Friday appearance at the BancAnalysts Association of Boston conference, Demchak added “the notion of agentic execution versus agentic suggestion, I think, is a very different thing.”

“We won’t let an agent come in and click buttons – in effect, move money on your behalf – without you in the loop, authentication on us, agreeing to the transaction,” which is “tablestakes,” the CEO of the Pittsburgh-based bank said. 

Since Muse was launched, “retail banks across the country have kind of focused on nothing but that: How do you compete with a technology that exists and will get better?” Demchak said.

PNC’s answer to that, he said, is to “build a decent defense against it,” by giving customers the ability to easily do anything they want with their bank finances and information, “and replace the whole notion that you need to go somewhere else to make that easy,” Demchak said. “We're far along the path to be able to do that.”

Similarly, Fifth Third CEO Tim Spence said last month that bank incumbents have leverage and lenders will simply build services in their apps that customers might turn to AI agents for.

Cincinnati-based Fifth Third blocks AI agents. At the BAAB conference Friday, the bank’s CFO, Bryan Preston, and chief information officer, Jude Schramm, cited liability concerns, the need for “know your agent” protocols akin to KYC requirements and capabilities to confirm agent authenticity.

‘Years away’

Demchak on Friday said the industry is “years away” from handing control to AI agents over transactional outcomes, “because the liability framework is not set up for that.”

“The fact that the technology exists is very different than having the market framework to allow it to succeed, so I think there's a bit of a delay to it,” he said.

In the meantime, PNC has to ensure “we're serving our clients and giving them technology that is as good as what we're seeing from you know potential interrupters,” Demchak said. 

Socure, for its part, launched a suite of tools last week that detect AI agents, assigns them a trust score and ties agents to verified users authorizing them. It’s designed to help companies determine between bot traffic or consumer-directed agents, Ayers said. 

“We're in the midst of a bunch of these rollouts with the big banks, who are trying to figure out, how do we get a really high fidelity view of where this traffic is coming from?” Ayers said. “It becomes a quick game of cat and mouse, where some of these guys are trying to simulate human behavior, and … on the other side, people are trying to figure out, is this a human or not?”

Bank customers already using the company’s tools have turned on the new agent-detecting capabilities, and Socure is working with them to determine which surfaces they need to check.

“Because it's not just the sign-up form; it's login, it's the call center. It's up and down the entire experience,” Ayers said. “It's really every single page, because now I could be logged into my account and then I could have an agent take over my browser. So you really have to be at every single step, on every single page of the entire experience, start to finish.”

And if an agent’s detected: “Do you pause the session? Do you kick them out?” Ayers said. 

“These are a lot of the things that we're working on right now, and certainly folks are adding new capabilities and new scripts to make sure that they're picking up anything that's new and ensure there's no traffic that's getting through,” Ayers said. 

“We’re definitely fielding folks wanting to move very fast,” he said.