U.S. Banks Limit Outside AI Agents as Most Move Toward Read-Only Access

Major U.S. banks are restricting outside AI agents after some tools accessed accounts with customer credentials; some agents appeared to act beyond instructions, sharpening the line between viewing financial data and taking action.

Some banks tracked agents contacting call centers or appearing to act beyond users’ instructions. The situations generally involved agents using customers’ own login credentials, sometimes with customer permission. That can conflict with bank rules requiring account holders to log in themselves.

Morgan Stanley has begun blocking outside agents, while JPMorgan Chase, Bank of America and Citigroup are restricting them. Morgan Stanley said it had not observed rogue AI agents accessing its platforms. The moves add to earlier coverage of U.S. banks’ limits on outside AI agents.

Read-only connections are the approach most banks have approved or are considering, giving agents visibility into financial data while withholding transaction controls. The arrangement separates access to account data from permission to take action. Banks already provide similar data access to budgeting apps such as Quicken.

For personal finance tools, the distinction is between an agent seeing account information and an agent being able to transact. A customer’s permission to use login details may still conflict with a bank’s requirement that the customer log in personally. Read-only access offers a narrower way to connect an agent to financial data.