For two years the tech industry has promised that AI agents will handle your errands, pay your bills and optimize your savings while you nap. America's biggest banks have now read that pitch and responded with the financial equivalent of a bouncer's raised palm. Turns out "let the robot move my money" is a sentence that makes compliance departments break out in hives.
Wall Street Locks the Side Door
Morgan Stanley is blocking outside AI agents from customer accounts, while JPMorgan Chase, Bank of America, Citi and Wells Fargo are restricting them, Axios reported on October 22. Citi says it doesn't permit third-party AI agents to log into customer accounts at all, though customers can still grant limited, read-only access through approved open banking providers.
Axios also reported that some agents appeared to take steps without user instructions. Reuters said it couldn't independently verify those incidents, and Morgan Stanley said it hadn't seen "rogue AI agents" on its platforms. In many of the cases described, customers had simply handed their login credentials to an agent, which is exactly the kind of access the banks never signed off on.
The Real Problem Is Identity, Not Intelligence
Strip away the sci-fi framing and the core issue is wonderfully mundane: a bank can see which credentials logged in, but it can't tell whether a human or an agent is behind them, or whether the customer actually approved what happened next. Bank of America's chief executive summed up the industry mood: "We strongly believe there should be a human in the loop to authorise money moving out of your account." Regulators have reportedly backed the banks privately, and the likely compromise is view-only access, where an agent can read your balance but can't pay your rent.
The angle worth stealing for any business is this: agents that borrow a human's password are a design smell, not a feature. Every company plugging AI into customer data is about to face the same question the banks just answered: what can the agent see, what can it do, and who is on the hook when it does something nobody asked for? There's also a less noble motive in the mix; banks reportedly worry agents could sweep idle deposits into higher-yield accounts elsewhere. Security and self-interest can point in the same direction.
Agentic AI isn't dead at the bank. It's just been told to keep its hands where everyone can see them.
Planning to let AI agents act inside your own site or customer portal? Talk to our team about scoping permissions and approvals before your users discover the hard way what "read-write" means.
Source: Axios