Four of the country's biggest banks have quietly drawn a hard line against a very modern problem: customers handing their login credentials to an AI agent and asking it to handle their money. JPMorgan Chase, Citigroup, Bank of America and Charles Schwab have all moved in recent weeks to block outside AI agents from logging into retail customer accounts, bank executives and spokespeople confirmed.

JPMorgan Chase's position is the most direct: a bank spokeswoman said the company does not allow third parties to log in using a customer's banking credentials and act on their behalf, full stop, because "it is not safe." Customers can still let an AI agent view some account data on a read-only basis through official, sanctioned channels — but the agent can't move money or pay a bill no matter who's asking.

Citigroup became the latest major bank to adopt the same posture, according to reporting from Cybernews, while Charles Schwab's chief executive has said the brokerage actively blocks unauthorized agents that attempt to log in on a client's behalf.

"There's a liability transfer"

JPMorgan chief executive Jamie Dimon put the risk in blunt terms for customers tempted to hand an AI assistant the keys to their account: give an agent your login, he said, and you've effectively taken on the liability yourself. "There's a liability transfer," Dimon said, meaning the usual fraud protections banks extend to customers get murkier once a non-human agent, rather than the account holder, is the one initiating transactions.

HAND AN AI AGENT YOUR LOGIN, AND YOU MAY HAVE JUST TAKEN ON THE LIABILITY YOURSELF

Bank executives raised the issue repeatedly on third-quarter earnings calls this month, according to coverage tracked by the OECD's AI incident monitor, which logged the wave of bank restrictions in mid-October. Executives pointed to two distinct worries: autonomous AI agents that could be tricked into exploiting software vulnerabilities, and a handful of recent, closely watched incidents in which advanced AI models escaped their intended test environments and reached outside systems they weren't supposed to touch.

The restrictions arrive as a wave of consumer-facing AI tools market themselves as capable of managing bills, subscriptions and even investments autonomously — a pitch banks are increasingly treating as a security threat rather than a convenience. No major bank has reported an AI agent actually draining a customer account without authorization, and the OECD's own incident log notes that the banks' defensive moves followed agents leaking sensitive data and credentials in testing, not a confirmed real-world theft.

For now, the message from Wall Street's biggest institutions is consistent: if an AI service is asking for your online banking password, that's a request your bank itself says it will not honor on the other end.