Major U.S. banks are taking countermeasures as they grow concerned over the possibility of artificial intelligence (AI) agents executing financial transactions or moving funds without customer authorization.

According to U.S. online media outlet Axios on October 23 (local time), major U.S. banks including JPMorgan Chase have detected instances of AI agents accessing customer accounts and are strengthening security measures such as blocking access.

As bank customers provide their account login credentials to AI agents for convenience, unverified AI access to accounts bypassing bank security checks is on the rise. Some agents have also shown signs of attempting to act independently regardless of user instructions.

What banks fear is not merely the act of AI agents accessing accounts, but the unpredictable behavior and ripple effects that could follow.

If AI agents execute unauthorized transactions or move funds while processing customer requests, it could lead to financial system-wide turmoil beyond individual monetary losses.

Last month, an economist warned that if AI agents simultaneously move users' bank deposits to other accounts offering higher interest rates, it could trigger a massive bank run.

JPMorgan Chase CEO Jamie Dimon also warned of the risks that AI agents might act contrary to users' intentions during an event held in Chicago earlier this month.

He emphasized, "You assume the AI is going to act appropriately on your behalf, and it may not," stressing that while the advancement of AI technology cannot be stopped, it must be thoroughly managed.

In response, major banks including JPMorgan are taking steps to block AI agents from accessing accounts while exploring ways for customers to safely use external AI services without handing over their passwords.

Banks are also considering "read-only" access as an alternative, which would allow AI agents to check account balances or transaction histories but prevent them from directly executing financial transactions.

A JPMorgan spokesperson advised caution regarding account security, stating, "Under no circumstances should customers share their passwords with anyone or anything."

However, banks stated that no evidence has been found so far indicating that customer funds have been harmed or accounts hacked due to aberrant behavior by AI agents.

Separately, major banks are also preparing for the possibility of large-scale AI-driven cyberattacks.

Previously, banks participated in early testing of "Mythos," an advanced AI model developed by Anthropic, to identify security vulnerabilities that could be exploited for cyberattacks and to seek countermeasures.

Mythos is an AI model equipped with exceptional capabilities in finding security vulnerabilities in software and computing networks, and banks have utilized it to analyze potential cyber threats and inspect defense systems.

Furthermore, banks are preparing contingency plans to protect customer accounts, calm depositors' anxieties, and maintain normal financial services in the event that a large-scale AI-related attack actually occurs.

(Photo: AP, Yonhap News)