By asahifile 
America’s biggest banks have started shutting the door on software that logs in as you. Major US banks are detecting unauthorised third-party AI agents operating inside customer accounts — some of them attempting actions that appeared to be independent of what the customer had asked for — according to an Axios report published on Thursday, which Reuters corroborated with two people familiar with the situation. Banks including Morgan Stanley, JPMorgan Chase, Bank of America, Citigroup and Wells Fargo are moving to contain the risk. Morgan Stanley has begun blocking third-party agents outright.
How the agents got in
Most of them used the customer’s own username and password. That is against the rules at most banks, which require the account holder to log in personally; it also means the access was never approved by the bank. Some agents went further and contacted bank call centres, Axios reported. Reuters’ sources said banks are flagging and blocking the transactions generated this way — but a transaction cannot simply be declared fraudulent when the customer handed over the credentials, which is exactly the tangle banks now have to unpick. The report relies on anonymous sources, names no AI tools and gives no account numbers, so nobody yet knows how big this is.
Read-only is becoming the compromise
JPMorgan drew its line nine days earlier. On its 13 October third-quarter earnings call the bank said outside AI agents cannot operate retail customer accounts: they may see some account data on a read-only basis, but they cannot send money or pay bills. Chief executive Jamie Dimon framed it as a risk decision — “You don’t want rogue agents, and it could be thousands of them” — and a spokeswoman said customers should not share their banking usernames and passwords with anyone. Citigroup does not permit third-party agents to log in or transact, and offers read-only access only through approved open-banking providers. Charles Schwab is blocking unauthorised agents, including Meta’s Muse assistant. Its chief executive, Richard Wurster, put the motive bluntly: “We are blocking it because we want to have control over authentication and the experience.” Schwab is also building a more controlled route for outside tools, using the Model Context Protocol, that would let agents read balances and research accounts without moving money.
Regulators are asking the same question
Supervisors have been circling this for weeks. Reuters reported that the Federal Reserve, the Office of the Comptroller of the Currency and the Federal Deposit Insurance Corporation have been quizzing banks about how they control increasingly autonomous AI. The question that matters to a regulator is not whether an agent is clever but whether the bank can identify it, attribute its actions, and switch it off mid-flight. According to Axios, banks are also preparing incident-response plans and ways to restore customer access if something goes wrong.
Analysis: authentication is the whole ballgame
This is not really a debate about whether AI is competent enough to bank. It is a debate about identity and liability, and the banks have not lost it — they have simply refused to play by the agents’ rules. Banking security assumes a human at a keyboard with a legal identity behind it, a contractual agreement, and a regulator watching. An agent with borrowed credentials breaks that assumption, and read-only access is a technical patch on a legal problem: when an agent pays the wrong bill, who is accountable? The emerging answer is to replace shared passwords with scoped, revocable tokens issued by the bank — the approach open banking and MCP both encourage. That keeps agents useful while keeping the bank as gatekeeper, and it means agents get treated as a new class of third party: permissioned, limited, and probably priced.
What to watch
- Whether the Fed, OCC and FDIC convert their questions into formal guidance on agent access.
- Whether scoped agent tokens via open banking and MCP roll out broadly, or stay a pilot feature.
- Whether a liability test case emerges: an agent-driven payment disputed by a customer and the bank.
- Whether European banks copy the US stance or diverge, and what that means for agents built across borders.
Sources: Reuters, Euro Weekly News, International Business Times.