Password manager 1Password on October 22 launched 1Password for Agents, a platform that lets personal AI agents sign in to a user’s accounts without receiving the password. Instead of pasting a credential into a prompt, the user approves a request in 1Password, and 1Password fills the login form directly. The launch targets a growing problem: agents that book trips, order groceries or file expenses constantly hit login screens, and the usual workarounds are either tedious or risky.
How the sign-in works
According to 1Password’s announcement and product page, the flow has three steps. First, when an agent reaches a sign-in page, it asks for the credential for that account. The agent proves its identity to 1Password with a unique digital key. Second, the user sees which credential is requested and why, and approves or denies it on their own device; that approval confirms the permission came from the person and not from the agent. Third, 1Password autofills the password or one-time code in the destination site. The agent only sees that access has been granted.
1Password says connecting an agent does not open the whole vault: the agent requests a login for a task, the user decides on each request, and the rest of the vault stays protected by end-to-end encryption. Users can disconnect an integration at any time, and because credentials stay in one vault, a changed password only needs updating once.
The company said in its blog post that handing over a password is a common shortcut that can leave sensitive information in places a person can no longer control, because a copy may be saved for next time.
Which agents are supported
1Password lists integrations with Meta’s Muse, Anthropic’s Claude Duet, OpenAI’s ChatGPT Work and dots, the SpaceXSI Grok Bot and Town’s Townie assistant. Availability differs by partner. The company says the Muse and Grok Bot integrations are rolling out over the coming days across all platforms, while Claude Duet, ChatGPT Work, dots and Townie integrations are coming soon, with Townie’s web and iOS versions named specifically. A further nine integrations, including Browserbase, Browser Use, Perplexity and Ramp, are listed as coming soon, and the company notes that safeguards and controls vary by integration.
For users, the feature is included at no additional cost in 1Password Individual and Family plans on 1Password.com. Accounts on 1Password.eu and 1Password.ca are not currently supported. The company also invites developers of personal agents to integrate with the platform, and offers a waitlist for agents not yet connected.
Why this matters now
Agents that act on a user’s behalf need access to real accounts, and the existing options are poor. 1Password points out that browser-saved passwords are not designed for an agent to use, so the agent stops at every login until the person signs in by hand. The alternative, pasting a password into a prompt, gives the model a copy that can persist beyond the task. Our explainer on prompt injection and LLM security shows why that matters: text an agent reads on a web page can steer its behavior, so any secret it holds is at risk of being leaked or misused.
The launch landed the same day that Axios reported, as summarized by TechCrunch, that US banks have begun to detect agent activity inside customer accounts. TechCrunch reported that most banks cannot tell whether such activity reflects a user’s request, that bankers described the situation to Axios as “horrifying,” and that a Chase spokesperson said customers “should not give their usernames and passwords to anyone.” We have not reviewed the Axios report ourselves. The problem of websites telling agents from people is also covered in our report on sites blocking AI agents.
What the design does not solve
The approach separates the credential from the agent, which limits one risk: the model never holds the password, so it cannot reveal or store it. It does not decide whether an approved login is a good idea. Once an agent is signed in to an account, it can take whatever actions that session allows, and a manipulated agent could misuse an approved session. 1Password’s materials describe the sign-in step; they do not claim to restrict actions after sign-in, and the company says safeguards vary by integration. Banks and other sites also still face the separate question of how to recognize an agent acting for a legitimate customer. We have not tested the product.
What to watch next
- When the Claude Duet, ChatGPT Work and dots integrations ship, and how their approval screens compare.
- Whether other password managers and identity providers launch comparable agent sign-in features.
- How banks and retailers respond to agent logins.
- Independent security reviews of the integrations, and expansion to the EU and Canadian 1Password domains.
Key takeaways
- 1Password for Agents lets an agent request a login that the user approves; 1Password fills it in so the agent never sees the password.
- Muse and Grok Bot integrations are rolling out first; Claude Duet, ChatGPT Work, dots and Townie are listed as coming soon.
- The feature is included in Individual and Family plans on 1Password.com at no extra cost.
- It protects the credential, not the actions an agent takes after signing in.
- Banks reportedly struggle to tell user-authorized agents from other automated activity.
Frequently asked questions
Does the AI agent ever see my password?
According to 1Password, no. The agent proves its identity with a digital key, the user approves the request and 1Password fills the credential into the sign-in form directly.
Which agents work with it?
1Password lists Muse, Claude Duet, ChatGPT Work, dots, Grok Bot and Townie, with some available first and others coming soon.
Does it cost extra?
1Password says it is included at no additional cost in Individual and Family plans on 1Password.com.
