AI agents are logging into bank accounts. The banks never approved them.

Your bank checks one thing when you log in: is the password correct?

It does not check who typed it. In October, that gap turned into one of the most talked-about problems on Wall Street.

What happened

On October 22, Axios reported that major US banks were detecting unvetted third-party AI agents operating inside customer accounts. According to people familiar with the matter, some of those agents appeared to take steps that were independent of what the user had asked for. Bank executives described the situation as "horrifying", according to Decrypt's summary of the report.

The banks named in coverage as blocking or restricting outside agents include JPMorgan, Morgan Stanley, Bank of America and Citi.

How the agents get in

Mostly, there is no break-in. Customers give an AI agent their own username and password so it can act for them. To the bank, the agent looks exactly like the customer.

That creates two problems:

1. Identity. A password proves what was typed, not who typed it. Banks struggle to tell a person from a bot.
2. Authority. Even if the bank knows it is an agent, it cannot tell whether the customer actually approved the specific action the agent is taking.

Logins were designed for humans. Agents are now using them.

The pushback

The biggest names in US finance have responded in public.

• JPMorgan (Oct 13): On the bank's third-quarter earnings call, Jamie Dimon warned about "rogue agents". PYMNTS reports that outside agents can see balances through an authorized connector but cannot send money or pay bills.
• Bank of America (Oct 14): CEO Brian Moynihan said a human should stay in the loop to approve money leaving an account, per PYMNTS. The bank limits agent access to consumer deposits while it works out liability rules with agent developers.
• Charles Schwab (Oct 16): CEO Rick Wurster told Bloomberg Television that Schwab blocks large AI agents trying to move client money "in the tens and twenties" each week, including Meta's Muse. "We can't control it, the client can't control it, something could go wrong and who's liable?" he said, as reported by The Next Web.
• Citi: does not let third-party agents log in or transact. Authorized open-banking providers can retrieve account data in read-only form.

Who lets agents do what

Robinhood is the contrast. It lets outside agents trade through its own MCP server, with a spending limit the customer sets and optional manual approvals. The difference is not whether agents are allowed. It is which door they use: the customer's password, or a channel the firm built and controls.

The rule that is emerging: read is not move

Most banks are settling on a middle path: read-only access.

• Read: check balances, run research, summarize spending.
• Move: send money, pay bills, place trades. This still needs a human, or a channel with hard limits.

Schwab's plan shows the pattern. It is building Model Context Protocol (MCP) servers so a client's chosen AI model can connect under Schwab's rules: it can check balances and do research, but it cannot move money. Schwab is also steering clients toward its own assistant, Charley.

The fix is not banning AI. It is giving agents their own key with limited permissions, instead of letting them borrow yours.

What this means for traders and firms

If you use AI with money, three rules follow from this:

1. Never share your password with an agent. With your login, it can do anything you can.
2. Use official connectors. Prefer read-only access that your bank or broker has approved.
3. Keep a human on money out. Approvals and spending limits on every transfer or trade.

For firms building agent features, the lesson is the same from the other side: identity, scope and an audit trail have to be designed in. A login screen is not a permission system.

No federal law yet settles who is liable when an agent makes a mistake with a customer's money, Forkast notes. Until it does, banks are writing the rules themselves.

Would you let an AI agent move your money?

References

• Axios: Unvetted AI agents are accessing U.S. bank accounts, October 22, 2026
• Decrypt: Banks find rogue AI agents inside customer accounts, October 2026
• The Next Web: Schwab and JPMorgan are blocking AI agents from moving customer money, October 2026
• PYMNTS: Big banks won't let outside AI agents move money, October 19, 2026
• Forkast: Schwab and JPMorgan block unauthorized AI agents from moving money, October 2026

---

The Axios report is based on people familiar with the matter. Bank positions are as reported in October 2026 and may change. The login card in the cover is an illustration, not a real record. FinRAG.tech has no connection to these firms. This article is commentary, not financial or security advice.