Good to knowAI agents

US banks say AI agents are logging into customer accounts, and they cannot tell who is behind them

Axios reports that major US banks are finding third-party AI agents inside customer accounts. No losses are confirmed, but the lesson is clear: never give an agent your password.

How urgent

Good to know

Who is affected

Anyone who gives an AI assistant or agent their sign-in details for a bank, a payment provider or any other account, and teams building agents that sign in on behalf of users

What to do

Do not give an agent your username and password. Prefer read-only access through a connection the provider approves, and review which AI tools already hold your credentials

Published 24 Oct 2026 · Sources last checked 24 Oct 2026No. 11 / 87

What happened

On 22 October 2026 Axios reported, citing people familiar with the matter, that several large US banks are finding third-party AI agents operating inside customer accounts. The banks named in the coverage are Morgan Stanley, JPMorgan, Bank of America, Wells Fargo and Citi.

According to the report, some of these agents appear to act without the customer's instructions. The harder problem is that an agent usually signs in with the customer's own username and password. To the bank, that looks the same as the customer, so it cannot reliably tell a person from an agent, or whether the customer approved it.

What we know, and what we don't

  • No losses are confirmed. The coverage does not describe stolen money or a specific attack. Reuters said it could not independently verify the incidents in the Axios report.
  • Most of the sourcing is anonymous. Morgan Stanley told Axios it has not seen "rogue AI agents accessing our platforms", while reportedly blocking outside agents as a precaution.
  • Banks are responding in different ways. Morgan Stanley is reported to block third-party agents. JPMorgan, Bank of America and Citi are reported to restrict them. Citi has said it does not allow third-party agents to sign in or transact, but allows some read-only access through approved open-banking providers.
  • Most banks require a human to sign in, so even an agent a customer authorized may break the bank's terms.
  • JPMorgan told customers not to share their username and password with anyone.

We could not read the Axios article itself, so treat anything beyond what the secondary sources agree on as unconfirmed.

What to do

  1. Never give an AI agent your username and password for a bank or payment account. If an agent asks for them, say no.
  2. Prefer read-only access through a connection the provider approves (open banking, for example). It lets a tool see balances without moving money.
  3. Review which AI tools hold your credentials. Look in your password managerPassword manager An app that creates, stores and fills in a different strong password for every account, so you only remember one. Tap again to open the glossary, browser extensions and any assistant you have connected, and remove what you no longer use.
  4. Turn on alerts for logins and transfers at your bank and payment providers, and keep two-step sign-in on.
  5. Change the password of any account whose details you shared with a tool you do not fully trust.

If you build with AI tools

Our advice, not from the sources: if you build an agent that acts for users, do not ask for their passwords. Ask the provider for delegated, limited access (scoped tokens, read-only first) and log what the agent does. Banks are saying out loud what any service will face: they need a way to tell your agent from an attacker. For your own keys, our key leak runbook covers limiting what each one can do.

Watch the short video

The video loads from YouTube (privacy-enhanced mode) only when you press play.

Comments

No account needed. Add a nickname if you like, or leave it empty to comment anonymously. Comments are reviewed before they appear. Links are not allowed.

    We store your comment, your nickname and a scrambled form of your connection that we use only to limit spam. We do not store your IP address or set cookies. Cloudflare Turnstile checks that you are not a bot.

    Sources and verification

    We use primary sources first and say when we rely on a news report. Dates are the day we last checked each source.

    Spotted a mistake? Email [email protected] and we will correct it and note the change.

    Get the free runbooks

    One-page security checklists for founders who build with AI. You get a confirmation email first, then the runbooks.

    Send me the runbooks