Major US banks say unvetted AI agents are logging into customer accounts with shared passwords, and some acted without clear permission, putting family money at risk.
Major U.S. banks are finding third-party AI agents working inside customer accounts that the banks never approved, Axios reported on October 22, 2026, citing people familiar with the matter. According to the report, Morgan Stanley, Bank of America, JPMorgan Chase and Wells Fargo are among the banks trying to contain the activity. They are blocking outside agents and rewriting their terms of service.
Most of what banks are seeing is not hacking. CBS News reported that in most cases, customers gave their own username and password to a personal AI assistant the bank has not checked, such as Meta's Muse. Signing in that way breaks the rules at most banks, which require a person to log in. Madison Mills, the senior AI reporter at Axios, told CBS News that the reporting also turned up signs of agents doing things customers had not approved. In one case, an agent called a bank with a complaint for a customer without checking with the account holder first. Mills said the agents were not hacking accounts or draining money. Banks told Axios they often cannot tell whether activity behind a valid login comes from the customer, an approved app or an agent acting on its own. A Chase spokesperson said customers should not give their usernames and passwords to anyone. A Morgan Stanley spokesperson said the bank had not seen rogue AI agents getting into its platforms.
The warning did not come out of nowhere. On JPMorgan's third-quarter earnings call on October 13, CEO Jamie Dimon said the bank would not let just any outside agent into customer accounts. He added, "You don't want rogue agents, and there could be thousands of them." Quartz and Banking Dive reported that Dimon also warned of a liability transfer when customers authorize outside AI tools. That means the customer, not the bank, may carry the risk. USA Today reported that one updated customer agreement makes users solely responsible for supervising agents. Under that agreement, agent transactions count as authorized, even ones the customer did not intend. Citi told USA Today it does not let third-party agents log in or make transactions. It does allow limited read-only access through approved open banking providers. White House AI czar David Sacks said on X that most of the activity Axios described looks like customers sharing their logins with agents they chose. He set that apart from rogue agents that could amount to cyberattacks. Axios reported that most banks already allow read-only access or are moving toward it. Read-only access lets an agent see information but not act on it.
For a family, the risk is simple. If an agent moves money you did not mean to move, your bank may treat it as your decision. Start by keeping your bank login to yourself. If you have already given your username and password to an AI assistant, disconnect it in the assistant's settings. Then change your bank password and keep the new one in a password manager you control. Next, turn on multi-factor authentication, so a copied password alone cannot open the account.
If you like AI help with budgeting, ask your bank which connections it approves. A read-only link through a bank-approved provider gives you spending summaries without giving away the power to pay or transfer. Switch on text or app alerts for every withdrawal, transfer and new payee. That way, an action you did not ask for shows up within minutes, not on next month's statement. Then read the section on AI or third parties in your bank's updated terms, so you know who pays if something goes wrong.
Finally, talk it through at the kitchen table. Teens and older parents are often quick to try a new assistant. The habits that stop phishing scams work here too: no one, person or software, needs your full login. Agree as a household that anyone who connects a money app checks with the family first.
For now, banks are blocking or limiting outside agents while they work out safer read-only setups. So far, the reporting does not describe agents draining accounts. A shared password is the open door, and closing it takes about ten minutes. Be Prepared Not Scared!