A recent query to Muse, the Meta Platforms Inc. personal agent: “Are you my agent in a legal sense?”

The initial response from the AI-powered bot: “Functionally yes, legally not quite.”

Wall Street is pressing technology companies to iron out exactly what “not quite” will mean in the future and who will pay if the technology runs amok.

Muse — responding to a prompt from the Bank Policy Institute, an industry lobby — elaborated that it doesn’t owe the user a fiduciary duty of care “the way a lawyer or financial adviser would.”

As consumers increasingly hand over their account credentials and credit-card numbers to these automatons, Wall Street banks fear their systems will soon be facing off against an army of rogue agents, which can steer trillions of dollars in and out of bank accounts, shop online, open new cards or switch to a cheaper mortgage. Because consumers have technically given these agents the go-ahead, banks want to make sure it’s clear who’s legally responsible when things go wrong.

“Before this technology scales to hundreds of millions of Americans, we need to get some fundamental protections right,” Cam Fowler, chief executive officer of Zelle parent Early Warning Services, which is owned by a consortium of banks, said in a statement. “That work needs to happen now.”

The question of liability is one of the biggest hangups for the financial industry, some of whose members have expressed increasing alarm about the rise of personal agents from Meta and OpenAI. For months, the biggest banks and payment firms had been working through the complexities needed to complete a so-called agentic transaction — where consumers could prompt AI to complete a task on their behalf.

Then in September, Meta released Muse, and the race was on — faster than many people expected.

Banks came out swinging. JPMorgan Chase & Co. Chief Executive Officer Jamie Dimon said his firm was blocking the agents on the consumer side. Others including Charles Schwab Corp., American Express Co. and more recently Royal Bank of Canada have taken similar steps, with many firms saying they want to first clarify who will be on the hook. Retailers such as Amazon.com Inc. have also begun prohibiting access.

Personal agents can offer a range of financial recommendations, including where to find the highest-yielding savings accounts and the best deals on insurance policies. Bank executives are particularly worried that the bots might one day be directed by their human overlords to move money autonomously, a concern that became even more acute after some of the most advanced models breached other companies.

Consumers are increasingly starting to use AI agents to send money. At JPMorgan, agentic consumer payment volume more than quadrupled since the introduction of Muse, even though the actual amount of money that moved in those transactions is relatively small, according to the company. By 2030, agent-executed volume could total $470 billion, Bloomberg Intelligence analysts Diksha Gera and Sena Ustuner wrote in a September report.

More From Bloomberg Intelligence

Agentic AI is poised to reshape how consumers choose banks before significantly disrupting deposit economics, with Oliver Wyman’s Aaron Fine viewing customer acquisition as the more immediate risk. Adoption is highly generational, with just 4% of consumers 55 and older interested in autonomous-agent transactions vs. 40% of younger consumers, limiting near-term disruption. Yield-seeking behavior also isn’t new, though agents could make moving excess deposits easier. Banks are restricting agent access given security, authorization and liability concerns.

– Herman Chan, senior industry analyst

Click here for the research.

More than 10 million people have used Muse in the US and Canada over its first six weeks, Meta CEO Mark Zuckerberg said during a conference call with analysts last month.

Some banks, while cautious, are more open to providing access to AI agents.

M&T Bank Corp., for example, has seen bots extracting customer information such as account balances, but aren’t moving money or asking to change products or services, Chief Financial Officer Daryl Bible said during a Nov. 5 conference.

“We’re closely watching it right now and if we see anything that we think is not appropriate, we will try to shut that down,” he said. “But right now we aren’t seeing anything that we’re concerned about.”

In addition to questions about liability, there are also concerns that people who deploy AI agents to help with one problem might create new ones.

For example, directing an agent to switch to a cheaper insurance policy might leave a consumer without the necessary coverage, or moving money to an account with a higher return without considering an upcoming rent payment could result in an overdraft fee.

Brian Shearer, a consumer protection attorney, said he’s most concerned about what happens when fraudsters start using AI agents to facilitate their attacks.

“If AI firms are collecting data from banks, it means that they will now have potentially the most sensitive and important information about a large group of people,” Shearer, director of competition and regulatory policy at the Vanderbilt Policy Accelerator, said in an interview. “I’m a little worried about the AI firms bumbling into collecting this information.”

A rule that would have governed how consumers can access and share their financial data with third parties was finalized by the Consumer Financial Protection Bureau during the Biden administration, prompting bank lobby groups to sue almost immediately. The Trump administration then scrapped it, leaving the issue largely unregulated just before the boom in agentic-driven transactions.

Today, banks are dealing with the consequences of their decision to fight, according to former CFPB Director Rohit Chopra.

“A few years ago, banks had the opportunity to be part of a system of consumer-permissioned data as part of a federal effort, and they heavily resisted it and fought tooth and nail to block it,” Chopra said. “Because the system never made that transition, we are now in a new era where consumers are needed to give their passwords to AI companies in order to move their money in a more methodical way.”

Written by: — With assistance from Yizhu Wang @Bloomberg