US banks detect unauthorized AI agent activity in customer accounts

According to Axios, large US banks are finding traces of third-party AI agents in customer accounts. Some of them are said to attempt actions that appear independent of the user's own instructions.

Banks are also regularly recording calls from agents to call centers and attempts to transfer money.

Journalists note that bankers are alarmed by the situation, and several banks are already preparing for worst-case scenarios and thinking about how to quickly protect or restore customer accounts if something goes wrong.

Most American banks are still generally opposed to agents making financial transactions on behalf of account owners. JPMorgan, for example, blocks services such as Muse and Instinct. Many other banks are doing the same, insisting that agents should have at most read-only permissions.

How institutions plan to tell authorized users apart from their agents remains unclear.