What banks are seeing

Several large US banks have found third-party AI agents logging into customer accounts, according to people familiar with the matter cited by Axios. In most of these cases the customer handed the agent their own login details and permission, which still conflicts with the rules of most banks that require people to log in to their own accounts. In other instances, banks tracked agents that contacted call centers or appeared to act independently of user instructions. Axios's sources said some institutions are already aware of unauthorized access. Whether those agents really acted on their own, rather than on a flawed reading of a request, has not been established.

Why the risk is growing

The concern is sharpest when an agent holds purchase authority or the ability to move funds, because a misread instruction can become a real financial transaction. Banking executives told Axios the industry has grave concerns about agents gaining access to customer accounts and taking steps that could affect user finances. Bank sources said their systems cannot always tell whether activity comes from a person or an AI tool, or whether that tool is following a legitimate instruction from the account holder. Axios described this difficulty in telling a human from an unauthorized bot as the main reason banks are forced to block the technology and rewrite their terms of service.

How banks are responding

Morgan Stanley has begun blocking third-party agents from its platforms, according to sources familiar with the decision. JPMorgan, Bank of America and Citi are also restricting agents, Axios reported. The Financial Times reported on 16 October that Citi does not allow outside AI assistants to log into customer accounts or make direct transactions through its digital platforms. Bank of America CEO Brian Moynihan has called for a human to approve money before it leaves an account. A Morgan Stanley spokesperson said the bank has not observed instances of rogue AI agents accessing its platforms, while a Chase spokesperson told customers not to share usernames and passwords with anyone.

Middle ground and liability

Most banks have authorized, or are moving toward, read-only access for agents, which would let them see financial information without taking action. Banks already grant similar access to third-party apps such as Quicken for budgeting. JPMorgan finance chief Jeremy Barnum told a quarterly conference that services relying on shared credentials are insecure, and the bank blocks such access where it can. Chase's terms of use reportedly require AI agents to identify themselves as programs and flag automated access, and make users responsible for errors or actions outside the authority they granted. Engadget, citing the US Electronic Fund Transfer Act, argued that customers who hand over their credentials are unlikely to get losses back. Industry executives say customers will keep wanting these tools, but for now the risks of being too permissive are too high.