1Password for Agents launches: AI agents complete sign-ins without seeing the password
1Password launched 1Password for Agents on 22 October 2026, a platform that lets personal AI agents complete sign-ins on a user's behalf without the agent ever seeing or storing the password.
1Password launched the 1Password for Agents platform on 22 October 2026, letting personal AI agents complete sign-ins on a user's behalf without the agent ever seeing or storing the password. The first integrations, Muse and Grok Bot, are rolling out in the days after launch — but several of the best-known agents will have to wait, and the feature is so far limited to consumer customers on 1Password.com.
How the mechanism works
The core idea is that the credential never reaches the agent or the model behind it. According to the press release, distributed via Business Wire, the sign-in works like this: when the agent arrives at a login form, it identifies itself to 1Password with a unique digital key. The user then sees which credential the agent needs — and why — and decides whether to approve the request. If the user approves, 1Password fills the credential directly into the form, rather than sending it to the agent (press release via AOL).
Access is also limited in time and scope. According to the company, as reported by Martin Cid Magazine, access is granted per session, restricted to the specific elements the user has approved, and it does not carry over to the next task. The user can disconnect an agent at any time. Passwords and multi-factor codes pass through a channel that 1Password controls, so they never pass through the model (Martin Cid Magazine).
The company's people stress that the human stays in the loop. "The user is involved in every decision," John Torrey, chief business officer at 1Password, told PCWorld. "Our customers decide which credentials an agent can use, for which task, and for how long. The agent cannot approve its own access" (PCWorld).
1Password's chief product officer Aashish Tripathi framed the thinking this way: "An agent should be able to do the job it has been asked to do, without getting permanent access to everything behind the account" (TNW).
What is actually available now
The rollout is gradual. Only the Muse and Grok Bot integrations are rolling out now, in the days after the announcement. Links for Claude Duet, ChatGPT Work and dots, and Townie are marked "coming soon" — with no date given (Martin Cid Magazine).
In addition, 1Password says several partners are working to bring support on board: Browserbase, Browser Use, Dana, Instinct, Kernel, Perplexity, Ramp, Tomo and Wajo, with "more integrations on the way" (The Verge).
Two names stand out by their absence: Google's Gemini Spark agent and Microsoft's Scout are not on the launch list (PCWorld).
Who gets to take part — and who is left out
The feature costs nothing extra, but applies only to Individual and Family plans on 1Password.com. That excludes everyone whose vault sits on 1Password.eu or 1Password.ca — the regional servers the company operates for European and Canadian customers. For European users, this means the platform is not yet available via the regional vaults (Martin Cid Magazine).
One source conflict is worth noting here: Engadget reported that pricing and availability were unknown, contrary to Martin Cid Magazine's details of a free add-on with a gradual rollout. The conflict is not resolved in the source material, but multiple sources confirm the gradual availability.
The background: from Claude extension to platform
The launch is not a sudden leap. 1Password built the platform on Secure Agentic Autofill, an early-access feature the company launched together with Browserbase in October 2025 (TNW). A Claude integration arrived in July of this year, before the broader agent platform was now officially launched (Engadget).
The press release also includes an endorsement from Nat Friedman, head of product at Meta Superintelligence Labs: "I have used 1Password since 2007, and I am very glad that it is now integrated directly into Muse. Muse is like an AI personal assistant, and 1Password gives it secure access to services, at your command" (press release via AOL).
What is verified — and what is not
The security architecture has so far been described only by the company itself, at a high level: a digital key, user approval and a closed autofill channel. No independent security analysis of the new mechanism exists in the source material.
There is, however, one independent test of a predecessor: PCWorld's reviewer tested the earlier 1Password for Claude extension at its launch this summer, and it "worked as advertised" — Claude Cowork signed in to the reviewer's FreshDirect account, shopped for groceries and even completed checkout. But that was the old extension; the new platform has not been tested hands-on (PCWorld).
There are also several details on which the sources disagree. Grok Bot's ownership is unclear: Engadget wrote "SpaceXAI Grok Bot", AOL and TNW "SpaceXSI Grok Bot", while the press release refers to it as "OpenAI, SpaceXSI Grok Bot" — the attribution cannot be resolved from the available material. Tripathi's title is also inconsistent: the press release, Engadget and TNW call him chief product officer (CPO), while The Verge describes him as chief executive officer. The best-documented title is CPO.
Why it matters
For anyone using AI agents for practical tasks — shopping, ordering, managing accounts — passwords have been one of the hardest obstacles: the agent either had to be handed the credential, or not do the job at all. 1Password for Agents proposes a third way, in which the agent completes the task without ever owning the key.
Open questions remain: how the mechanism behaves in real-world practice, when and whether it reaches business customers and the European and Canadian regions, and who Grok Bot actually belongs to. Until independent tests exist, the claims about its security rest on the company's own word.