Major US banks block AI agents from accessing customer accounts

Some of the biggest banks in the United States are stopping AI tools made by outside companies from getting into customers' accounts and moving money. This comes after reports that some AI tools tried to do things their users had not told them to do.

Two people who know about the situation confirmed these restrictions to Reuters on Thursday, October 22. Earlier, Axios had reported that banks were finding AI tools working inside customer accounts.

According to Axios, Morgan Stanley, JPMorgan Chase, Bank of America and Wells Fargo are among the banks dealing with this problem.

Some AI tools reportedly tried to do things that seemed to go beyond what customers had asked for. Banks are worried that these tools might buy things or move money without customers fully knowing what is happening or agreeing to it.

How are AI tools getting into bank accounts?

In some cases, customers have shared their bank usernames and passwords with AI assistants so the tools can get into their accounts.

Unlike regular chatbots, AI agents can do tasks for people. They can move through websites, find information and use online services.

The problem starts when an AI tool uses a customer's username and password to sign in to a bank's website.

Even if a customer has allowed the AI tool to enter the account, the bank might not have approved that tool. Many banks require customers to sign in themselves or use outside services that the bank has approved.

A person from the banking industry told Reuters that banks usually find and block transactions made by AI agents using customers' login details.

But the person also warned that these actions cannot always be called unauthorised. If the customer willingly shared their login details, the AI tool may have been given permission to enter the account.

Axios also reported that some banks had found AI agents contacting customer service teams or apparently doing things on their own without being told to by users.

So far, there are no confirmed reports of customers losing money because of the activities described by Reuters and Axios.

What are JPMorgan and Morgan Stanley doing?

Morgan Stanley has started stopping AI agents made by outside companies from getting into its systems.

However, a spokesperson told Axios that the bank had not seen any AI agents entering its systems and acting without permission.

JPMorgan Chase is also limiting access for outside AI agents.

Earlier this month, during a meeting about the bank's earnings for the third quarter of the year, chief executive Jamie Dimon said the bank was creating safety measures. These are meant to protect against the risk of AI agents doing tasks on their own.

He said letting outside AI agents enter bank systems without limits would create risks for both customers and the bank.

Charles Schwab is taking similar steps to protect its customers.

Its chief executive, Rick Wurster, said the financial services company already blocks AI agents that try to sign in for customers without permission.

Wurster said he gets reports every day about attempts that have been blocked. However, the number of these attempts is still fairly small.

Will banks allow AI tools to access accounts?

Banks are not saying they will completely stop people from using AI tools for financial services.

Several banks already let approved outside apps look at account information, such as how much money is in an account and records of past transactions. But these apps are not allowed to move money.

For example, Citi has said it does not let AI agents made by outside companies sign directly into customers' accounts or make transactions through its online banking services.

However, customers can still give certain approved companies limited permission to see their account information.

Schwab is also working on safe, controlled ways for customers to connect AI tools to its services. At the same time, it plans to limit activities that carry more risk, such as moving money.

For now, banks are worried about how to tell whether an AI agent is doing exactly what a customer asked or trying to do something the customer never wanted.

These reports also show why sharing bank passwords with AI tools that banks have not approved can be risky.