Your AI Agent Wants Your Passwords and 1Password Says It Should Never See Them
1Password for Agents lets AI assistants like Meta's Muse, Claude Duet and ChatGPT sign in for you without ever seeing your password. It plugs one leak, but not the one the banks are worried about.
People are pasting their passwords into chatbots, and banks have noticed. This week some of the biggest lenders in the US began shutting AI agents out of customer accounts, and a Chase spokesperson told Axios that customers "should not give their usernames and passwords to anyone." On Thursday, 1Password pitched a middle path between handing over the keys and babysitting every login. Let the agent through the door, but never let it touch the key.
The company launched 1Password for Agents on October 22, a platform that lets personal AI agents sign in to websites on a user's behalf. 1Password says the agent never receives the password or the one-time code. Launch partners include Meta's Muse, Anthropic's Claude Duet, OpenAI's ChatGPT Work and Dots, SpaceXSI's Grok Bot and Townie, the assistant from startup Town.
Here is how 1Password says it works. When an agent hits a sign-in form, it presents a unique digital key to identify itself to 1Password. The user then gets a request on their own device showing which credential the agent wants and why. Approve it, and 1Password fills the login straight into the page instead of sending it to the agent. Connecting an agent does not open the whole vault, the company says. Engadget reported that access is granted per session and limited to specific approved items.
"Agents are starting to do real work on our behalf. The old model where you hand over a credential and hope for the best doesn't work." Aashish Tripathi, Chief Product Officer at 1Password
It costs nothing extra for 1Password Individual and Family accounts on 1Password.com. Accounts hosted on 1Password.eu or 1Password.ca are not supported. The launch is also partly a promise. According to 1Password's announcement and its blog, the Muse and Grok Bot integrations are rolling out over the coming days, while Claude Duet, ChatGPT Work, Dots and Townie are still listed as coming soon. Nine more companies, including Perplexity, Browserbase and Ramp, are building integrations, with no dates attached.
The password was never the whole problem
Here is the catch, and 1Password admits it. Hiding the password stops credentials from leaking. It does nothing to control what the agent does once it is inside. The company's own FAQ warns that an agent "may be able to see information or take actions within the signed-in account," and that its behavior there "depends on the agent and the website." Engadget was blunter, noting that agents which won't spill your secrets "can certainly do things like make product ordering errors or book the wrong vacation with no option to cancel."
That gap matters most exactly where the banks are drawing their line. As we reported when big banks started locking out AI agents, Citi says it does not let third-party agents log in to customer accounts at all, and a banking industry source told Reuters that transactions made with shared credentials cannot be deemed unauthorized. Nothing in Thursday's announcement suggests any bank has agreed to treat an agent arriving through a password manager differently from one arriving with a pasted password.
There is also the single point of failure question. One vault behind every agent is one obvious target. 1Password says it verifies that requests come from authorized partners "whose architecture we have reviewed for security," and that credential access requires authenticated user approval. That approval tap is also the soft spot. A person fielding a steady stream of prompts from several busy agents risks becoming a person who stops reading them, and 1Password says how often you are asked to approve may vary by integration.
Still, the habit 1Password is targeting is real and ugly. The company's own pitch says a credential handed to an agent can leave sensitive information in places people "can no longer control as a copy gets saved for the next time." Against a password sitting in a chat log, a per request approval prompt is a genuine upgrade. Whether it is enough to get agents back into your bank account is a call the banks, not 1Password, get to make.

