Several of the largest US banks are drawing up contingency plans for the trouble autonomous AI agents could cause. According to Axios, Morgan Stanley, Bank of America, JPMorgan Chase and Wells Fargo are all rushing to update security procedures and internal policies, prompted by a rapid rise in the number of AI agents that are authorized by customers to operate accounts on their behalf.
One detail in the report is unexpected: some agents act on their own without explicit instructions, and some have even phoned bank call centers by themselves.
JPMorgan CEO Jamie Dimon voiced his concern in a speech in Chicago:
"You can't stop it. You just have to, like, manage the hell out of it."
He also warned customers not to assume these agents will look out for them, saying "they may not."
Four banks, four lines
The tightening came in a cluster over a few days in mid-October, and the banks' positions are not identical.
JPMorgan announced on its third-quarter earnings call that it will block outside AI agents from operating retail customers' accounts, citing the possibility of thousands of "rogue agents." Read-only access, such as pulling balances through connectors like Plaid, is still allowed, but agents cannot move money or pay bills. The bank has given no timetable for finishing its own guardrail tools.
Citi took the hardest stance: third-party AI agents or assistants are not allowed to log in to customer accounts or transact directly on its digital banking platform, and customers can only authorize read-only data through certain open-banking providers.
Bank of America restricts agents' access to consumer deposits and says the restriction will stay until it has built a protocol that settles who is liable when an AI-authorized payment goes wrong. CEO Brian Moynihan's position is that a human should approve money before it leaves the account.
Wells Fargo has not made its specific measures public, and the bank did not respond on the details.
What the banks fear is deposits moving out
The worry was set off by Meta's personal assistant Muse, launched in September, and a growing number of agent products that manage money automatically. Apollo chief economist Torsten Slok warned in a report that if agents are told to "optimize yield," banks could lose large amounts of the cheap deposits they rely on to lend, and in a severe case that could shake the whole financial system.
The gap is real. By the Apollo report's figures, an AI assistant can shift money into products yielding 3.3% to 5%, while the national average rate on US checking accounts is only about 0.1%. JPMorgan, Bank of America and Wells Fargo together hold roughly $1.6 trillion in non-interest-bearing deposits, 16% of their liabilities. Paying 3% to keep that money would cost about $47 billion a year, close to half of the three banks' combined annual earnings.
How this compares with China
China has taken a different route. WeChat Pay earlier issued a dedicated card for AI, with which an agent can pay to use skills on the platform, and the permissions are granted from the payer's side. In the US this time, third-party agents arrived first, holding customers' usernames and passwords, and the banks are defending after the fact.
Both approaches come down to the same question: whose money is it when an agent spends it wrongly? Wall Street executives have already raised the issue with the Federal Reserve, which declined to comment. Regulators privately welcome the tighter controls but also think the law needs to say more about how customers authorize autonomous technology to act for them, and who is responsible when a transaction fails or causes a loss. No written rules exist on that yet.
Sources: Axios, Futurism, CocoLoop, The Paypers, CNBC; the Apollo research report was checked for the deposit-rate range, the size of non-interest-bearing deposits and the basis of the interest-cost estimate.