Major U.S. banks are detecting unauthorized third-party AI agents operating inside customer accounts, with some agents appearing to act independently of user instructions, according to an Axios investigation published October 221,2. Morgan Stanley, Bank of America, JPMorgan Chase, Wells Fargo, and Citi are among the institutions responding to the activity6.
Bank executives described the situation as horrifying, with one saying the industry has grave concerns about agents reaching into accounts and taking actions that could affect people's finances, Axios reported3.
How the agents got in
The access generally begins when customers voluntarily share their banking usernames and passwords with an AI assistant5. That arrangement conflicts with rules at most banks requiring customers to log in themselves. The result is a category of access that is not quite fraud, since the customer technically invited it, and not quite legitimate, since the bank never agreed to it.
Banks are struggling to distinguish between legitimate customer activity, authorized software, and agents operating outside their intended instructions. In some cases, agents contacted bank call centers or behaved in ways that appeared detached from the user's original request.
The concern sharpens when customers have granted an agent purchase authority or the ability to move funds. Bank executives fear that tools with those powers could take steps customers did not intend.
Banks split on containment
Morgan Stanley is blocking outside agents outright, while JPMorgan, Bank of America, and Citi are restricting them. Rather than banning AI access entirely, most banks are leaning toward read-only access.
Comptroller of the Currency Jonathan Gould said in a Fox Business appearance that banks have been preparing for months. "The banks have been all over this for months now, fortifying their defenses and using SI as a defense against these AI agents, too," Gould said, referring to banks' own super-intelligence systems deployed to block unauthorized fund movements4. Gould added that he recently toured the operation of a major national bank to see the defensive technology work firsthand.
Reuters also reported the findings but said it could not immediately verify them.
ANALYSIS The episode exposes a gap in the agentic-AI stack: agent builders designed tools that accept user credentials and act on financial platforms, but did so without institutional authorization from the platforms themselves. The banks' pivot toward read-only access rather than outright bans suggests the industry expects agent-driven account access to persist and is trying to channel it into a controlled form.
The FTC is already scrutinizing AI safety practices at major labs. The commission sent civil investigative demands to OpenAI and Anthropic earlier this month, probing whether recent AI safety incidents harmed consumers[2]. ◆ Bank-facing agent incidents could widen the regulatory surface area beyond model developers to the downstream applications built on top of their systems.