Banks are shutting out AI agents that log in with customers' passwords

America's biggest banks have found AI agents inside their customers' accounts, and most of them were never approved by the bank. The agents log in with the customer's own username and password, sometimes with the customer's blessing, which breaks the rule at most banks that people log in to their own accounts. Some have rung call centres. Some appear to act on their own.^[1]

Morgan Stanley has begun blocking third-party agents from its platforms, and JPMorgan, Bank of America and Citi are restricting them, Axios reported on 22 October. Wells Fargo is among the banks working to contain the risk. Executives worry most about agents that customers have allowed to make purchases or move money, and several banks told Axios they cannot reliably tell an agent the customer authorised from one that has gone rogue.^[1]

Why the banks are pulling up the drawbridge

JPMorgan Chase drew its line on its third-quarter earnings call on 13 October. Chief financial officer Jeremy Barnum said services that move customers' money using shared banking credentials are being shut out. "It's not secure, and so to the extent that we're able to block those, we are blocking them," he said, "to protect customers and protect ourselves."^[2] Chief executive Jamie Dimon was blunter about the scale: "You don't want rogue agents, and it could be thousands of them."^[3]

A JPMorgan spokeswoman told CNBC: "Because it is not safe, we do not permit third parties to access accounts."^[3] Morgan Stanley said it had not observed "rogue AI agents accessing our platforms".^[1] The banks are rewriting their terms of service and considering read-only access, which would let an agent see account data without being able to act on it, the same arrangement many already give budgeting apps.^[1]

The Australian catch: your password, your loss

Australian customers face a sharper problem than a blocked login. Under the ePayments Code, which most banks here have signed, a customer whose unauthorised transaction stems from choosing to share their passcode is generally "not entitled to a refund".^[4] Typing a banking password into an AI assistant or agent so it can pay bills or move money is exactly that kind of sharing. If the agent is compromised, misreads an instruction or is hijacked, the loss may sit with the customer, not the bank.

Australian banks are not refusing agents outright. They are building their own and backing schemes that check who is acting. Commonwealth Bank launched Companion, an AI agent inside its app, in May.^[6] CBA and Westpac are trialling Mastercard's Agent Pay, which requires a biometric check such as Face ID for each purchase and uses tokens that tie every transaction to a specific authorised agent.^[5] Surin Fernando of Mastercard Australasia said: "Often retailers don't know it's an agent making the transaction."^[5]

The industry's lobby group has the issue on its radar. Australian Banking Association chief executive Simon Birmingham warned a parliamentary inquiry into AI in early October about agentic AI intersecting with the financial system.^[7]

What customers and businesses should do

Do not give a banking username and password to any AI tool, browser agent or app that is not your bank's own. Use the bank's app features or schemes the bank has signed up to, such as Agent Pay, where the bank can see that an agent is acting and the customer confirms each payment. Businesses should check whether staff have connected AI assistants to company bank accounts or payment platforms, and switch on alerts for new devices and logins.

The banks' move comes in a week when security leaders warned that AI is finding software flaws faster than organisations can fix them. Bushletter's report on that warning and the AI and security series cover the wider picture.