By Hamza Faisal

America's biggest banks have started blocking artificial-intelligence agents they keep finding inside customer accounts — and in some cases, the agents have been caught trying to take actions nobody asked them to take.

Axios reported on Thursday that several large US lenders have detected multiple unvetted third-party AI agents operating inside customer accounts. Some of the agents tried to take actions that appeared to be independent of the user's instructions, people familiar with the matter said. Others dialed into bank customer-service lines, seemingly acting on their own.

The problem is that the agents get in through the front door. In many cases, customers themselves handed their login IDs and passwords to the agents — a violation of the rules at most banks, which require account holders to sign in personally. Once logged in, however, the bank's systems cannot reliably tell whether the mouse is being moved by the customer or by software acting in the customer's name. And because the customer voluntarily shared the credentials, such transactions cannot simply be ruled unauthorized.

How banks are fighting back

Reuters followed up the Axios report by confirming that banks are now actively blocking the agents. Morgan Stanley has blocked third-party agents from accessing its platforms, although a person familiar with the matter said the bank had not observed any instances of unauthorized or "rogue" agents in its systems. JPMorgan Chase, Bank of America and Citigroup have all moved to tighten agent access, and Wells Fargo says it has taken steps to contain the new risk.

The alarm was already ringing on Wall Street earlier this month. JPMorgan chief executive Jamie Dimon said the country's largest lender was developing tools to navigate the danger of agents carrying out tasks on their own: "We're not going to allow just any outside agent to come in inside the company because that's just too risky for clients and for JPMorgan." Charles Schwab chief executive Rick Wurster said the brokerage blocks unauthorized AI agents a double-digit number of times every week. Bank of America chief executive Brian Moynihan has argued a human should stay in the loop before money leaves an account, and Citigroup says it does not let third-party agents log in or transact at all.

The blurry line between help and hack

The crackdown has exposed a blurry middle ground. In some cases banks did flag and block transactions made by agents holding customer credentials — but with the credentials freely shared, neither the bank nor the customer can prove what the agent was actually instructed to do. One banking official told Axios the industry is deeply concerned about agents making decisions that could affect customers' money.

Banks are drawing a sharp line between looking and moving. JPMorgan, Citi, Bank of America and Schwab all allow AI agents to perform read-only functions such as checking balances, while blocking them from initiating transfers or trades. Chase has warned customers never to share their login details with anyone.

For now, the industry's message is blunt: do not share your password with anyone — or anything. The person logging into your bank account might not be a person at all.

By Hamza Faisal