Banks are blocking AI agents. But it’s just an opening move

Ask around and it won’t take long to find someone relying on Meta’s Muse for some chores. The personal AI agent became the No. 1 free app in the U.S. Apple App Store within two weeks of launching.

In exchange for relinquishing control of their errands, the personal AI agent shops, sends emails, books appointments and more for consumers. But if the apps and websites Muse visits block the agent, Muse is about as useful as a house key after the locks are changed. And some businesses already are.

In the same two weeks, consumers who asked Muse to shop on Amazon for them hit a pop-up: “Continued access by an unauthorized AI agent violates Amazon’s Conditions of Use.” 

Amazon moved early but it’s far from alone. A number of big banks have already banned external AI agents from logging into customers’ accounts and taking on tasks like moving money because they deem them as unacceptable risks, including the biggest bank of all, JPMorgan Chase.

“We have to make sure we protect our customer in the right kind of way,” CEO Jamie Dimon said on the bank’s third-quarter earnings call. “So we're building tools to do that, too. In the meantime, we're not going to allow just any outside agent to come in inside the company because it's too risky for the clients and for JPMorgan." 

At Money20/20, PNC CEO Bill Demchak said agent-only transactions inside bank accounts will never happen; banks won’t cover losses from tech’s experiments. “Tech companies, they’re cool. They can do whatever they want and break stuff, particularly if we’re willing to pay the bill,” Demchak said. “But we’re not willing to pay the bill.”

Bank of America, Citigroup and Charles Schwab have also blocked or temporarily restricted third-party agents. Others, like Chime, are seeking to offer a middle ground. The different approaches represent what the next big fight in banking is over: What AI agents can and can’t do on behalf of bank customers. 

If it sounds familiar, it should. For more than a decade, the industry has fought over how consumers share banking data with other apps like Venmo or Credit Karma. The matter is hardly settled. The clearest result: Plaid now pays banks like Chase for data access.

Likewise, blocking AI agents is just the opening act as companies decide how to identify agents and who is liable for what when something inevitably goes awry.

Already, some banks are writing their terms down. For instance, Wells Fargo updated its terms and conditions to make customers who provide their login credentials or otherwise authorize an AI agent to act on their behalf on the hook for rogue agents, and starting Nov. 24, customers remain liable until they can prove they revoked access.  

Chase’s terms of use page now includes this language, “When you use, allow, enable, or cause the deployment of an AI Agent, you are responsible for any actions or errors that may result from the Agent’s incorrect interpretation or implementation of your instructions, if it performs unintended actions or exceeds the authority you gave it.”

Not every bank will draw the line in the same place.

“An agentic process without guardrails or a clear regulatory perimeter means the short term will be messy,” said Jane Barratt, chief advocacy officer at MX, in a Yours in Finance interview. “There will be push and pull.”

At a time when every big tech company is hustling to build the most useful agent that takes care of consumers’ tasks, companies are making snuffing one out look as easy as blowing out a candle. But it isn’t that straightforward. Some unvetted AI agents are still getting into bank accounts, according to Axios.

The companies trying to block AI agents have their stated reasons: They’re protecting consumers. But the message is a bit tangled when considering how some of those same companies, like Chase, are building their own agents. It doesn’t mean the banks are wrong. The story is just complicated. They also want to defend their customer relationships.

Plenty of people don’t want an agent in their bank account at all for now. “I have AI do repetitive tasks for me at work and use it fairly frequently,” one Reddit user wrote. “But I’ve never once had a reason to give it my bank account username/password. Nothing I do in my bank account even requires that sort of automation.”

Banks have a valid point, too.

“The security and safety and soundness of financial institutions is a really critical aspect of whether or not the whole system works and doesn’t descend into chaos,” Atomic CEO and Co-Founder Jordan Wright said in a Yours in Finance interview.

Still, blocking isn’t an enduring strategy for technology that isn’t vanishing. The question is how far to let agents in. 

As Wright sees it, the middle ground could mean only sharing read-only information with agents. Safe-to-spend estimates have been unreliable in the past and AI could help project a customer’s finances to show whether they can afford the next expense.

“You don’t know what else is coming up,” Wright said. “It’s very hard to understand all of that. Being able to compile that for a consumer and say ‘these are all the things that are coming up that you might not be thinking about right now. Are you sure you are safe to spend this?’ People have tried safe-to-spend numbers in the past but they have always been faulty.”

Others, like Chime, are seeking something in between. Chime offers customers an in-house agent, Jade, and also says it is building secure connections to general-purpose AI agents.

"We don't believe that a single agent will be best suited to manage every area of your life,” said Chris Britt, co-founder and CEO, on Chime’s third-quarter earnings call. “Researching a major purchase or planning a trip is one thing. In more complex, regulated categories like healthcare and finance, we believe that trusted specialist agents will win.”

If specialist agents win, so does whatever company offers them. 

As Wright sees it, customers will consolidate. As agents handle more tasks, people will go all-in on a single provider that offers them everything they need to manage their finances. 

“The next year is going to cement relationships that are going to be more concentrated than they’ve been in the last 10-20 years as financial institutions, fintechs and horizontal players battle it out for the best experience for their customers,” Wright wrote on LinkedIn.

Only time will reveal what approaches endure. But it’s already clear that consumers, not blocks,  will decide.