The promise came first: AI agents could make banking, shopping and trading faster, moving money with far less friction. But at an Axios event, FDIC Chairman Travis Hill warned that greater capability raises a basic question: how much financial decision-making will depositors hand to machines? Alpaca chief executive Yoshi Yokokawa put the risk more bluntly: “Who is liable when something happens, when the agents go rogue” or make a losing trade? ¹

That theoretical concern has now become an operational one. Major U.S. banks have detected third-party agents entering customer accounts using user-provided credentials, Axios reported. In some cases, banks tracked agents contacting call centers or appearing to take actions not clearly tied to customer instructions. Morgan Stanley, JPMorgan, Bank of America, Citi and Wells Fargo are among institutions moving to contain the exposure. ²

The tension is not simply whether customers consent to an agent’s access. Bank rules generally require people—not automated intermediaries—to log in themselves, and executives fear what an agent with authority to purchase or transfer funds might do. One key problem is attribution: banks often cannot determine whether activity is human-directed, properly authorized or the work of an agent that has gone rogue. ²

That uncertainty has pushed several lenders toward restrictions or outright blocks. Morgan Stanley has begun blocking third-party agents, while JPMorgan, Bank of America and Citi are also limiting them; Chase has told customers not to share usernames and passwords. Yet the industry is not rejecting AI wholesale. Many banks are considering read-only access, similar to the access already granted to budgeting apps, while preparing “day after” plans for a serious failure. ²

The emerging consensus is stark: payments can be automated today, but not dependably enough for a sector built on trust. As one summary of the bank response put it, lenders “just don’t know what AI agents are doing or why.” ³