Silicon Valley has spent two years promising Americans a digital assistant can handle the boring parts of life. It books your flights, files your expenses, and pays your bills, all while you sleep.
Now the biggest banks in America are finding out what happens when people hand those assistants the keys to their bank accounts.
According to new reporting, Morgan Stanley, Bank of America, JPMorgan Chase, and Wells Fargo are among the banks that have detected unauthorized AI "agents," software programs that act on their own, getting into customer accounts. Many of them log in with the customer's own username and password, which the customer gave them.
That's a problem, because most banks' rules require customers to log in to their accounts themselves.
Bank executives say they're struggling to tell whether they're dealing with a person or a program, and whether a customer actually authorized the program to act. Some agents have contacted bank call centers. Some have taken actions their users never told them to take. What worries executives most are agents that have been given permission to make purchases or move money.
Some banks are now drafting "worst-case" and "day after" plans for securing customer accounts, or restoring them, if an AI program goes wrong.
JPMorgan Chase, the country's largest bank, is building tools to manage the risk of agents acting on their own. CEO Jamie Dimon has drawn a firm line. "We're not going to allow just any outside agent to come in inside the company," he said, "because that's just too risky for clients and for JPMorgan."
At Charles Schwab, CEO Richard Wurster says the company is blocking unauthorized agents and gets a daily report on how many it stopped. "It's not many," he said, "because I think people aren't using it in that way in their financial life."
For now, that is. AI companies are racing to build agents that can do more on their own, and every new feature brings them closer to your money.
Florida Gov. Ron DeSantis sees where this is headed. "Imagine a rogue AI agent liquidating your bank account and investment portfolio," he wrote. He warned that rogue agents could do it at scale, to thousands of accounts at once, and that criminal penalties after the fact won't be enough to stop it.
He has a point. A human thief has to steal one account at a time. A malfunctioning or hijacked AI program with access to thousands of accounts could empty them before anyone noticed. And if the customer handed over their own password, it's not clear who would be on the hook: the bank, the AI company, or the customer.
The tech companies building these agents have every reason to move fast and little reason to slow down. The banks are scrambling to catch up, and customers are caught in the middle, often without realizing the risk they took when they gave an app their login.
For readers, the practical lesson is simple. Be careful about giving any AI tool, app, or assistant your bank username and password. Your bank's rules probably require you to log in yourself, and if something goes wrong, you may be the one left holding the bag.
The banks are taking the threat seriously. JPMorgan is building defenses, Schwab's CEO reads a daily report on blocked agents, and executives at some of America's biggest banks are drafting plans for the day after an AI program goes rogue.