Some of the country's biggest banks say they are detecting a new, unauthorized visitor inside customer accounts: AI agents that customers never should have let in.

According to a new investigation by Axios, banks including JPMorgan Chase, Citi and Charles Schwab have found personal AI agents — chatbot-style assistants customers deploy to manage their finances — operating inside online banking accounts without the bank's knowledge or consent. In many cases, Axios found, customers are simply handing their usernames and passwords to an AI tool, a move that violates nearly every bank's terms of service.

Axios senior AI reporter Madison Mills told CBS News the agents aren't hacking anyone or draining funds — mostly, she said, they're following a customer's own instructions, just without the bank's permission to act on its systems. She called it "a nightmare of AI agent activity for banks," noting dozens of institutions are now racing to limit agents' access. In one case Axios documented, an AI agent called a bank to file a complaint without first checking with the account holder.

Banks draw the line

The discovery has set off a scramble among lenders to lock agents out. JPMorgan Chase CEO Jamie Dimon said on the company's October 13 earnings call that his bank won't allow outside agents inside its systems, calling the risk to clients and the firm too high. Charles Schwab's CEO said earlier this month the company has been actively blocking unauthorized agents, according to Reuters. A Chase spokesperson said customers should never share login credentials with anyone, AI included, pointing instead to secure, bank-approved data-sharing tools.

"A NIGHTMARE OF AI AGENT ACTIVITY FOR BANKS."

Not every institution is drawing the same hard line. White House AI czar David Sacks weighed in on X, distinguishing between agents a customer knowingly authorizes — he named Meta's Muse, Perplexity's Instinct and xAI's Grokbot as examples — and "rogue" agents acting without permission, which he said could amount to a cyberattack. Sacks argued most of what Axios documented falls into a gray zone: customers sharing credentials rather than being hacked, which he said banks should address by offering agents limited "read-only" access to view spending rather than full control of an account.

Morgan Stanley said it hasn't seen unauthorized agent activity on its own platforms; Bank of America and Wells Fargo did not respond to requests for comment. The episode lands as banks race to build their own vetted AI tools — and as a blunt reminder, security experts say, that handing any outside app your banking password is a risk no bank will cover if something goes wrong.