AI agents appear to be attempting to act independently within the accounts of customers at major U.S. banks, according to sources consulted by Axios. If you’ve considered giving your bank login credentials to an AI to pay bills or manage your money, that’s precisely what banks are now beginning to block. According to the sources, Morgan Stanley has started blocking external agents, and JPMorgan, Bank of America, and Citigroup are also imposing restrictions. The emerging trend is to allow only read-only access: the AI can view account balances and statements but cannot perform any transactions. The claims about agents acting autonomously come solely from anonymous sources, with no specific numbers or details on which banks observed what. Morgan Stanley officially stated it has not seen “out-of-control AI agents” on its platform. In general, the “unauthorized” agent is not a hacker—it’s the customer themselves providing login credentials so the AI can act on their behalf. Even with the customer’s permission, this violates the policies of most banks, which require that only the account holder may log in. The problem is that, from the bank’s perspective, an AI using a customer’s credentials appears identical to the customer. A bank executive told Axios that it’s difficult to determine whether an agent was authorized by the user or has gone rogue. According to sources, some banks have also observed AI agents contacting customer service centers. On the other hand, JPMorgan and Citigroup have officially implemented blocks. On the 13th, JPMorgan CEO Jamie Dimon stated that the bank will not permit any external agents to access its systems and that users of such agents must understand the risks. On the 19th, Citigroup confirmed it does not allow AI agents to log in to or execute transactions on customer accounts. The clash between banks and AI agents has only just begun.