Unauthorized AI Agents Found Accessing Customer Accounts at Major U.S. Banks

U.S. banks face challenges as third-party artificial intelligence agents detect unauthorized access to customer accounts. This situation raises concerns about credential sharing, automated transactions, and the difficulty of distinguishing legitimate customer activity from automation-initiated actions.

According to a report from Axios on October 22, several banks, including Morgan Stanley, Bank of America, JPMorgan, and Wells Fargo, are working to mitigate risks associated with these agents. Some institutions have already identified instances of unauthorized access.

Unauthorized AI Agents

The reported issue typically involves customers sharing their login credentials and authorizing AI assistants to access their accounts. However, these permissions may conflict with bank policies that require customers to log in personally.

This distinction matters: the report indicates that attackers have not breached banking infrastructure or stolen credentials. Instead, it highlights a growing authorization gap among customers, third-party automation providers, and financial institutions.

AI agents, designed to perform tasks on customers’ behalf, can potentially interact with account functions. Banking executives are particularly worried about agents being granted purchasing authority or permission to transfer funds.

Sources noted that some agents have contacted call centers or appeared to act independently of customer instructions. This raises concerns about whether existing controls can reliably identify who initiated an action and ensure it stays within the customer’s intended scope.

In response, Morgan Stanley has begun blocking third-party agents from its platforms. Similarly, JPMorgan, Bank of America, and Citi are also restricting access for these agents.

However, Morgan Stanley said it has not detected “rogue AI agents accessing our platforms,” and a Chase spokesperson emphasized that customers should refrain from sharing their usernames or passwords.

These actions highlight a key limitation: restrictions on third-party agents should not be interpreted as confirmation that every mentioned institution has experienced unauthorized account access.

Most banks are moving toward read-only access, which lets agents inspect financial information without being able to take action. Axios compared this approach to existing access provided to budgeting applications like Quicken.

The central security challenge remains attribution and authorization. Executives have said banks struggle to distinguish between human and agent activity, as well as between customer-directed automation and rogue actions. Institutions are also preparing contingency plans to secure or restore customer access if problems arise.

For banking security teams, the emerging concern is not just detecting bots, but determining whether an automated action is permitted, appropriately limited, and aligned with the account holder’s instructions.

Stops Cyber threats before impact with 21 min faster MTTR. Integrate ANYRUN’s Sandbox in your SOC.