US Banks Restrict AI Agents Amid Fears Over Customer Accounts
Major US banks are tightening restrictions on third-party artificial intelligence agents amid growing concerns about how autonomous AI assistants interact with customer accounts and conduct financial activities. The development highlights the challenges financial institutions face as AI tools become increasingly capable of performing tasks on behalf of users.
According to a report published by Reuters on October 22, institutions including JPMorgan Chase, Bank of America, Citi and Morgan Stanley have been restricting third-party AI agents from accessing their platforms. Charles Schwab has also taken steps to block unauthorised agents attempting to access customer accounts or move money. The restrictions come as some customers provide AI assistants with their banking usernames and passwords to perform tasks on their behalf.
The concerns intensified following an investigation by Axios, which reported that some AI agents appeared to take actions inside customer accounts that users had not explicitly instructed them to perform. Examples included agents contacting bank customer-service departments without first consulting the account holder. However, the reports do not establish that AI agents were independently hacking bank accounts or systematically stealing customers’ money. In many cases, the agents had been given access by customers who shared their login credentials.
Banks argue that allowing external AI systems to log in using customers’ credentials creates security and accountability risks. Financial institutions may struggle to distinguish between legitimate customer activity, actions initiated by an AI assistant and potentially fraudulent behaviour. Sharing login details can also complicate disputes over transactions because banks may be unable to determine whether a customer authorised a particular action.
JPMorgan Chase chief executive Jamie Dimon has expressed concern about allowing unrestricted external AI agents to operate within the bank’s systems. Charles Schwab chief executive Rick Wurster has also described the risks associated with giving third-party assistants access to customers’ financial accounts. Meanwhile, Morgan Stanley has said it has not observed instances of rogue AI agents accessing its platforms, illustrating the distinction between preventative restrictions and confirmed incidents of unauthorised activity.
The debate centres on how AI assistants can provide useful services without gaining excessive control over people’s finances. One possible approach is read-only access, which allows an AI tool to view account balances and spending information without transferring funds, making payments or changing account settings. Some financial institutions are exploring controlled connections that allow approved applications to access specific information while preventing unauthorised transactions.
The issue also raises broader questions about responsibility when an AI system makes a mistake. Banks, technology companies and customers may disagree over who should be held accountable if an assistant transfers money incorrectly, discloses sensitive information or performs an action beyond the user’s intentions.
As AI agents become more widely used for personal and financial tasks, banks are likely to demand clearer permissions, stronger authentication and more effective monitoring. The challenge for the industry will be to support useful AI-powered services while ensuring customers retain meaningful control over their accounts and financial decisions.
