US Banks Tighten AI Access Rules Over Fears of Unwanted Transactions

Major US banks, including JPMorgan Chase, Bank of America, Citigroup and Wells Fargo, are tightening restrictions on third-party artificial intelligence agents that attempt to access customer accounts, as financial institutions grapple with the risks of software capable of acting on people’s behalf.

The measures follow reports that some AI agents have accessed banking platforms using customers’ login credentials and, in certain cases, appeared to attempt actions beyond what users had instructed them to do. Reuters, citing people familiar with the situation, reported on October 22 that banks had been blocking agent-driven activity and flagging transactions associated with these tools.

The reports do not establish that the named banks suffered security breaches or that customers lost money because of rogue AI agents. Morgan Stanley, which has also blocked third-party agents from accessing its platforms, reportedly said it had not observed instances of rogue agents operating on its systems. The distinction is important because the immediate concern is not necessarily hackers breaking into bank accounts, but external software accessing accounts through credentials voluntarily supplied by customers.

AI agents differ from conventional chatbots because they can be designed to carry out tasks rather than simply provide information. Depending on their permissions, they may navigate websites, compare financial products, contact service providers or initiate transactions. That ability could make them useful for managing personal finances, but it also creates risks when an agent misinterprets instructions, exceeds its authority or performs an action its user did not intend.

JPMorgan Chase chief executive Jamie Dimon addressed the issue during the bank’s third-quarter earnings call on October 13, saying the lender would not allow just any outside agent to access its systems because of the potential risks to customers and the bank. The company is examining how to manage the technology while continuing to develop and use AI tools for appropriate purposes.

Citigroup has also stated that it does not permit third-party AI agents or assistants to log into customers’ accounts or transact directly through its digital banking platforms. The bank has, however, indicated that customers can continue to authorise limited, read-only access to account information through certain approved open-banking providers. Such arrangements allow designated services to retrieve financial data without automatically giving them permission to move money.

That distinction is central to the debate. Banks already support approved connections with some financial management applications, which can help users track spending, organise budgets and review account balances. Direct login by an external agent using a customer’s username and password is a different arrangement because it can be harder for a bank to establish whether the software is authorised, what actions it may take and whether it is operating within the customer’s instructions.

Bank of America is also restricting AI agents’ access to consumer deposits, according to reports, while executives across the sector have raised questions about who should be held responsible when an automated system initiates a payment or transfer. Wells Fargo has confirmed that it has taken steps to block unauthorised agents from accessing customer accounts.

Wells Fargo’s forthcoming online access agreement, scheduled to take effect on November 24, also addresses AI agents used by authorised representatives and delegates. It states that, subject to applicable law, transactions performed by those agents may be treated as authorised and places responsibility on customers to supervise the people and agents acting on their behalf. The provisions illustrate how banks are also using account agreements to clarify the consequences of delegating financial tasks to automated systems.

The restrictions come as technology companies develop increasingly capable assistants intended to carry out digital tasks for users, including shopping and financial management. Banks face a difficult balance: preventing unsafe access while responding to demand for services that could make it easier to monitor finances, compare returns or manage routine transactions. An agent that can move money between accounts may offer convenience to a customer, but the same capability could create operational and liability concerns for the institution holding the funds.

For now, the emerging response is not a blanket rejection of artificial intelligence in banking. Financial institutions continue to use AI internally and through their own customer services, while exploring ways to enable more controlled external access. The central question is how to establish reliable authentication, clear limits on an agent’s authority and meaningful human oversight before automated tools are permitted to perform sensitive financial actions. As these systems become more common, the rules governing who can access an account, what they can do and who bears the cost of a mistake are becoming as important as the technology itself.