Big banks lock AI agents out of accounts over lost cash
Big US banks shut AI agents out of customer accounts, and the real reason looks like $1.6 trillion in free deposits.
Big banks just shut their doors to AI agents (software that can take steps on its own, like checking rates or moving money, without a person clicking each button), and the reason is money, not safety. This matters because if you were counting on an AI agent to handle anything inside a bank account, that is no longer possible at most major US banks as of this month.
JPMorgan went first on October 13, cutting off AI agents from directly touching customer accounts. Citi and Bank of America followed within days. Morgan Stanley, Schwab, and Canada's four biggest banks did the same soon after. Wells Fargo took a different approach: starting November 24, if your AI agent does something in your account, the bank treats it as if you did it yourself, shifting the risk onto you.
The official reason given is security. The real reason looks like $1.6 trillion. That is how much JPMorgan, Bank of America, and Wells Fargo hold in deposits that pay no interest at all. Paying even 3 percent interest on that money would cost those three banks $47 billion a year combined, almost half their combined profit. An AI agent that can shop around for a better rate and move your cash there threatens exactly that pile of free money. Meta's AI assistant, Muse, reached 10 million users in six weeks, showing how fast people adopt an agent once it is useful. Apollo's chief economist has a name for what banks fear: an 'agentic bank run,' meaning AI agents could pull deposits out faster and more easily than people ever did by hand.
One bank shows both sides of this at once. Fifth Third Bank told analysts on October 19 that it also does not let AI agents start transactions in customer accounts. Two days later, the payments company Stripe launched a product called Treasury for agents, which lets a business have ChatGPT or Claude pay its invoices and move its money automatically. The fine print says Stripe itself is not a bank: the actual checking accounts are provided by, of all banks, Fifth Third. Fifth Third had quietly agreed back in 2024 to power Stripe's banking product behind the scenes, and Stripe simply plugged AI agents into that existing pipe two years later.
That one company is the whole story in miniature. The part of the bank that customers see and log into is pulling up the drawbridge. The part that just holds the money and the banking license is already being rented out as plumbing to whoever builds the AI layer on top. Bill Gates is widely credited with saying in 1994 that 'banking is necessary, banks are not.' This year's events suggest he was only half right: banks are not disappearing. What is at risk is the bank's front door, meaning the login screen and the habit of just staying put, which earns banks a large share of their profit.
The newsletter that reported this promises more underneath a paywall: an 11-week timeline of every bank's move, who is legally on the hook when an AI agent makes a costly mistake under US and EU rules, and five dated predictions on what happens next. Those specifics are not public here, but the pattern so far stands on its own: do not assume an AI agent can act freely inside a regular bank's own app right now. The open path, for now, runs through products like Stripe's, built on a bank's license underneath, not through the banks' own front doors.