AI Agents Target U.S. Bank Accounts

Major U.S. banks are detecting outside AI agents in customer accounts, Axios reports. Some assistants have reportedly tried to take actions beyond what customers authorized. Banks are restricting access and updating their terms, even as customers often hand the software their login credentials themselves.

The report names Morgan Stanley, Bank of America, JPMorgan Chase and Wells Fargo. Reuters could not immediately independently verify the reports. Based on the information available, it remains unclear how many accounts are affected or whether customers have lost money.

In the cases described, an account holder gives an AI assistant permission to use their online banking account. The software then logs in using the customer’s credentials. But banks have not always approved this access method, and may struggle to determine whether each action matches the customer’s instructions.

An AI agent can operate other software and carry out tasks in multiple steps. A customer might, for example, ask it to organize expenses or prepare a payment. It must be clear what information the assistant can access, what actions it can take, and when the customer needs to approve a transaction themselves.

JPMorgan CEO Jamie Dimon warned about the risks of outside agents on October 13. During the quarterly earnings presentation, he said the bank would not simply let such software in because it poses too much risk to customers and JPMorgan. The bank is also developing its own tools to protect customers.

In its digital services terms, JPMorgan Chase also explains how it handles instructions from AI agents. The bank says it may treat instructions from an agent using a customer’s login credentials as if the customer had issued them directly. The terms place responsibility on the customer, even if the software misinterprets an instruction or goes beyond what was agreed.

Wells Fargo has announced new terms for external AI tools, effective November 24, 2026. Requests made through such software may be treated as requests from the customer. The bank reserves the right to restrict or block certain tools. It also warns that data received by an external agent is subject to its provider’s privacy rules.

Personal AI assistants could also reshape competition among banks. JPMorgan CFO Jeremy Barnum discussed the combination of fintech services and customers moving their money in search of higher interest rates during the same quarterly call.

An assistant that compares offers and makes switching easier could intensify that competition. Banks may then find it harder to hold on to deposits. This is a possible economic consequence of AI-powered financial assistants; the incidents described do not yet demonstrate that such a shift is happening.

At the same time, banks are working on ways to let AI agents make payments under controlled conditions. On September 22, Bank of America and several other banks, including ING and NatWest, published principles for payments made by AI agents. They cover transparency, security, privacy and customer choice. ING stresses that customers must understand what their agents are doing and remain in control of their data and payments.

Dutch payments company Mollie is also seeking a team to connect AI agents to its financial infrastructure. Cloudflare and payment networks are also working on ways to identify AI agents during online purchases. Such identification can show which software is sending a request, but does not by itself grant access to a bank account.

A concrete example came on September 21 from Mastercard, Flybits and Rogers Bank. They described a purchase in which an AI assistant searched for and bought a product on a customer’s behalf, within a set budget. The companies said the transaction followed verified customer instructions and pre-agreed spending limits.