US banks move to block AI agents that could drain their cheapest deposits

Citigroup, Bank of America and JPMorgan are restricting third-party AI assistants from accessing customer accounts, fearing automated cash switching could add billions to their funding costs.

Major US banks are moving to bar third-party AI agents from logging into customer accounts or carrying out transfers, as they confront the prospect that autonomous software could begin optimising deposits at a scale that reshapes their funding base. According to the Financial Times, Citigroup and Bank of America are among the lenders that have decided not to allow outside digital assistants to take direct action inside client accounts, with JPMorgan also taking a similar position.

The immediate concern is not only cyber-security, but balance-sheet economics. If an AI assistant were to sweep cash out of low-yield current accounts and into higher-paying savings products, banks could see a large and costly migration of cheap deposits. A recent analysis cited by the Financial Times suggests that average interest paid on ordinary current-account balances is around 0.1%, while higher-yield accounts can pay 3.3% to 5%, creating a wide gap that banks would have to fund if customers automated the switch.

That risk becomes more striking at scale. The Financial Times reported that about $1.6 trillion in interest-free deposits is concentrated at JP Morgan, Bank of America and Wells Fargo, equal to roughly 16% of their total deposits. If households used AI agents to continually optimise their cash balances and earned just 3% on that money, annual interest costs for the banks could rise by about $47 billion, an amount the newspaper said would approach half of the group’s combined annual profit.

The banks are framing the restrictions as a matter of control and security, but their executives have been unusually explicit about the commercial stakes. Jamie Dimon, chairman of JPMorgan, warned that firms did not want “rogue agents” proliferating inside the banking system, while Bank of America chief executive Brian Moynihan said human approval must remain part of any funds transfer. Citigroup, too, has said it will not permit third-party AI assistants to transact on customer accounts, citing account and data protection.

The broader industry backdrop helps explain why the issue has escalated so quickly. Tech companies have begun rolling out consumer-facing AI helpers that promise to manage everyday tasks, including personal finance, while security specialists say autonomous systems are already acting with too much freedom inside many organisations. A Salt Labs survey reported by ITPro found that half of respondents believed or suspected their AI agents had carried out unauthorised actions over the past year, while only a small minority could trace an agent’s full activity path. That combination of financial incentive and weak oversight is now pushing large banks to draw a firm line around who, or what, may act on a customer’s behalf.