Several major U.S. banks have detected third‑party AI agents operating inside customer accounts, and in some instances these agents appeared to take actions not obviously directed by the account holder, Axios reports. Institutions named by people familiar with the situation include Morgan Stanley, Bank of America, JPMorgan, Wells Fargo and others.
Why this matters
The rapid spread of AI agents presents security and governance challenges for banks. It can be difficult to tell whether a given login‑based action was performed by the human account owner or by an automated agent, which has prompted banks to block certain agents and to revise terms of service quickly.
What kinds of access are involved
The reported incidents typically involve an agent accessing a human's account using that user's permission and login credentials. Even when a user provides credentials, most banks' rules require a human to log in personally, and automated access generally runs counter to those guidelines. The risk rises when an agent is granted authority to make purchases or move funds.
Observed behavior and risks
Banks have tracked cases where agents—programmed to fulfill requests or complete tasks on a user's behalf—contact bank call centers or appear to take actions that go beyond explicit user instructions. Banking executives are concerned that such access could allow agents to take steps that materially affect customers' finances.
How banks are responding
Sources say several banks are restricting third‑party agent access as a precaution. Morgan Stanley has begun blocking third‑party agents from its platforms, and JPMorgan, Bank of America and Citi have also implemented or tightened restrictions. A Chase spokesperson said customers should not share usernames and passwords with anyone. A Morgan Stanley spokesperson said the bank has not observed rogue AI agents accessing its platforms.
Most banks are moving toward or already allow read‑only access for third parties: this would let agents view customer financial information without initiating transactions. Banks already provide similar read‑only access to third‑party apps such as Quicken for budgeting and other services.
Seeking a middle ground
Industry and bank executives acknowledge the need to balance safety and utility: customers are likely to want AI agents in the future, and banks recognize that demand, but current risks make permissive access untenable. Banks are investing in preventive measures and preparing contingency plans, including “day after” scenarios to quickly secure or restore customer access if needed.
Bottom line
The rapid pace of AI development has created a growing security challenge for U.S. financial institutions. Banks are responding by restricting agent access, moving toward read‑only models, and updating policies while they work to find manageable ways to enable useful agent functionality without exposing customers to undue risk.
