Major U.S. banks restrict AI agents’ account access, Axios reports

Customers’ permission is not enough to get an AI assistant through the bank’s door. Major U.S. banks are tightening account-access rules after some reportedly detected agents logging in with credentials supplied by customers. Axios reported the restrictions on October 22, according to Gizmodo’s coverage, describing a security response that separates reading financial information from taking action.

The login can look legitimate

The reported catalyst was not necessarily a stolen password or an agent breaking into an account. Some customers apparently gave agents their own login information and permission to use it. Axios said that arrangement still conflicts with most banks’ guidelines requiring people to log into their own accounts.

Recognizing that activity is another problem. An anonymous bank executive told Axios that banks generally cannot draw a clear line between an agent’s actions and a user’s. A successful login can therefore leave a basic question unanswered: whether the customer or software acting for them is operating the account.

The incidents need an important qualification. Gadget Review’s account says Reuters could not independently verify the specific cases described by Axios. A Morgan Stanley spokesperson also said the bank had not observed instances of “rogue AI agents accessing our platforms.” That statement challenges any suggestion that every bank tightening access has itself detected such activity.

Banks tighten the rules around outside agents

Morgan Stanley, Bank of America and Wells Fargo are updating their terms of service to prevent agents from taking autonomous actions inside customer accounts, Axios reported. Gadget Review’s account of the same reporting also names JPMorgan Chase and Citi among institutions taking steps to limit third-party agent access. These are accounts of one underlying Axios report, not independent confirmation of the incidents.

JPMorgan’s resistance was already public earlier this month. On an earnings call, CEO Jamie Dimon rejected unrestricted entry by outside agents, citing risks to both customers and the bank. His objection was about which systems the bank would admit, rather than treating a customer’s willingness to use an assistant as sufficient authorization.

We’re not going to allow just any… outside agent to come in inside the company, because it’s just too risky for clients and for JPMorgan

Jamie Dimon, JPMorgan Chase CEO, on an earnings call earlier in October, as quoted by Gizmodo

Reading a balance remains a different permission

The response is not a blanket rejection of every connection between an AI tool and financial data. Many banks have limited agents to read-only access, Gizmodo reported. That lets an assistant review information without permission to move funds between accounts or buy stocks. The restriction separates access to information from control over the assets it describes.

• JPMorgan permits limited, view-only information retrieval through intermediaries such as Plaid. That route does not authorize agents to transfer money or pay bills, according to Gadget Review.
• Citi says third-party AI agents cannot log into customer accounts or transact directly on its digital banking platforms. It allows certain read-only access through approved providers that share banking data.

The next move is not necessarily a permanent ban. Axios described banks acknowledging that customer use of agents will probably grow and that learning to work with them may be preferable to excluding them altogether. For now, the approved routes described by JPMorgan and Citi stop short of transaction authority. Any broader accommodation would have to go beyond the read-only access those banks currently permit.