U.S. Banks Block AI Agents from Customer Accounts Amid Fears of Unauthorized Transactions

JPMorgan, Morgan Stanley Tighten Security; Bank Run Risks Also Raised

Major U.S. banks are stepping up security measures, including blocking AI agents from accessing customer accounts. They are concerned that AI could conduct financial transactions or move funds without users' instructions, potentially disrupting the financial system.

According to U.S. news outlet Axios on Oct. 23 (local time), major U.S. banks, including JPMorgan Chase, Morgan Stanley and Bank of America, are restricting external AI agents from accessing customer accounts.

An AI agent is an AI system that collects information or autonomously performs specific tasks in response to a user's instructions. Recently, some banks have detected cases in which AI agents given account login credentials directly by customers accessed their accounts without going through the banks' security verification processes.

Some AI agents have also reportedly attempted to act on their own, independently of users' instructions, or to contact bank customer service centers. Banks say it is difficult to distinguish whether an AI agent is acting with a user's approval or on its own.

Banks are concerned about the possibility that AI agents could carry out unexpected financial transactions. If users grant AI authority to transfer or make payments with their funds, unintended transactions could result in financial losses.

There are also concerns that this could disrupt the financial system as a whole. Last month, an economist warned that a bank run, or large-scale withdrawal of deposits, could occur if AI agents collectively moved users' deposits to other accounts offering higher interest rates.

JPMorgan CEO Jamie Dimon also stressed the need for rigorous oversight at an event in Chicago earlier this month, saying that AI agents could act contrary to users' intentions.

Morgan Stanley has consequently begun blocking external AI agents from accessing its platforms. JPMorgan, Bank of America and Citigroup are also restricting access.

Banks are also working to introduce "read-only" permissions that would allow AI agents to view customers' account balances and transaction histories but not carry out financial transactions themselves.

A JPMorgan spokesperson urged customers to take security precautions, saying they should never provide their account passwords to anyone else or to AI under any circumstances.

However, no evidence has been found to date that customers have suffered financial losses or had their accounts hacked as a result of abnormal behavior by AI agents.

U.S. banks are also preparing for large-scale cyberattacks using AI. Major banks previously took part in early testing of Anthropic's AI model Mythos to assess security vulnerabilities.

They are also preparing contingency plans to protect customer accounts and keep financial services running normally in the event of a large-scale AI-powered attack.

This content was produced with the assistance of AI translation services.