U.S. banks are detecting unauthorized third-party AI agents accessing customer accounts, raising concerns about automated financial activity that institutions cannot reliably distinguish from human behavior.
Some agents have appeared to act independently of customer instructions. The report identifies Morgan Stanley, Bank of America, JPMorgan and Wells Fargo among institutions working to contain agent-related risks.
However, it does not establish that every named bank experienced unauthorized access or that confirmed losses resulted from the activity. In many reported cases, customers supplied their login credentials and permitted an AI assistant to access their accounts.
Major US Banks Detect Unauthorized AI Agents
That permission does not necessarily satisfy bank policies, which generally require customers to log in rather than delegate access through shared credentials. This distinction separates the reported activity from a conventional account takeover involving stolen passwords.
An agent can possess valid credentials and customer approval while still operating outside the financial institution’s authorized access arrangements. The potential consequences become more serious when customers grant purchasing authority or permission to move funds.
Banking executives expressed concern about what agents could do after accessing authenticated accounts, particularly when their actions could affect customer finances.
Sources also described agents contacting bank call centers or taking actions beyond user instructions. The report does not disclose specific transactions, affected customer totals, agent vendors, or technical evidence establishing how those apparently independent actions occurred.
The central security challenge is attribution: banks struggle to determine whether activity originates from a human, an approved assistant or an agent acting outside its intended instructions.
Executives told Axios that distinguishing between different agents and determining customer authorization remain difficult.
Morgan Stanley has begun blocking third-party agents. JPMorgan, Bank of America and Citi are also restricting agent access. A Chase spokesperson warned customers not to share usernames and passwords.
Morgan Stanley provided an important qualification, saying it had not observed “rogue AI agents accessing our platforms.” That statement underscores the difference between restricting unapproved automation and confirming malicious autonomous account compromise.
Most banks have authorized, or are moving toward, read-only access that lets third-party services view financial information without executing actions. Axios compared this approach with access granted to budgeting applications such as Quicken.
Banks are also revising service terms and preparing contingency plans to secure or restore customer access if incidents occur.
The emerging challenge is accommodating customer demand for AI assistance without treating possession of login credentials as sufficient authorization for unrestricted automated financial activity.
For now, restrictions reflect unresolved questions about identity, delegated authority, and control over account actions.
Join 16,000+ SOC teams using ANY.RUN to streamline threat investigations and reduce manual effort. Explore for your team