Banks lock out AI agents as 1Password offers a key

Handing your bank password to an AI agent just stopped working.

Some of the biggest US banks are blocking or restricting AI agents that sign in with customer credentials, Axios reported on Oct. 22, per Engadget and Gizmodo. The same day, 1Password announced its answer to the problem those agents created.

Banks found agents already inside

According to the Axios report, some banks detected AI agents operating in customer accounts using login details the customers had supplied. Banks also tracked agents contacting call centers without specific permission, Engadget reports, citing unnamed sources at several banks.

Morgan Stanley, Bank of America, JPMorgan and Wells Fargo made changes to contain the risk, Engadget says. Some blocked third-party agents outright, others restricted them, and many are trying to hold agents to read-only access: they can see balances but not move money. Gizmodo reports that Morgan Stanley, Bank of America and Wells Fargo are updating their terms to stop agents such as Meta's Muse from acting on their own in accounts.

A Chase spokesperson told Axios customers should not share usernames and passwords with anyone, AI agents included, per Engadget.

Muse is the agent most named. Meta said on Oct. 21 that it passed 10 million US downloads in its first 40 days, Gizmodo reports.

The password was always the weak link

The trouble is how agents get in.

An agent using your password looks exactly like you. Gizmodo reports an unnamed bank executive saying banks often cannot tell agent activity from the customer's own. And if an agent moves money wrongly, Engadget argues a refund is unlikely, because a bank would probably treat it as something you chose to do.

1Password for Agents, announced Oct. 22, changes who holds the key. When an agent reaches a sign-in form, it identifies itself to 1Password with its own digital key. You see which login it wants and why, and approve or decline on your own device. 1Password then fills the credential straight into the form, so the agent never receives it.

"The old model where you hand over a credential and hope for the best doesn't work," said Aashish Tripathi, 1Password's chief product officer.

The launch partners are the agents banks are worried about: Meta's Muse, Anthropic's Claude Duet, OpenAI's ChatGPT Work and dots, and Grok Bot, plus Perplexity, Ramp, Browserbase and Browser Use. "Claude Duet asks before it signs in anywhere," said Ami Vora, Anthropic's chief product officer.

What to do this week

If you have given a browser agent your bank login, take it back. Change the password. Banks are rewriting their terms against it, and you may carry the loss alone.

Yes, 1Password's approach is better. It still does not make an agent welcome at a bank whose terms say a human logs in. Approval on your phone keeps your password private. It does not change who the bank thinks is at the keyboard.

If you build agents, stop designing around stored credentials. The direction is clear: approvals per sign-in, read-only by default, and front doors made for agents, like the gateway Walmart opened this week and World ID's proof that a human sent the agent.

What to watch

1Password has not given a price or a date for general availability. Watch for both, and for the first bank to say it will accept a delegated sign-in instead of blocking it.

Until then, keep your agents out of your accounts. Read-only or nothing.