1Password for Agents Lets AI Assistants Log In Without Passwords

1Password has launched 1Password for Agents, a new platform that lets personal AI assistants — including Meta's Muse, SpaceXSI's Grok Bot, and Town's Townie, with more agents on the way — sign into websites on a user's behalf using a stored vault credential, without the agent ever receiving or seeing the actual password. The company announced the platform on October 22, 2026, in a blog post by Chief Business Officer John Torrey. Each sign-in still requires explicit user approval, and 1Password says "the safeguards and controls vary by integration" from one AI partner to the next.

The launch extends a model 1Password first shipped in July 2026, when it let Anthropic's Claude access saved logins in the browser without exposing the underlying secret to the model. 1Password for Agents turns that same approach into a standing platform that any supported personal AI agent can plug into.

What 1Password for Agents actually is

1Password for Agents is a set of integrations between 1Password's password manager and a growing list of personal AI agents. According to 1Password's announcement, when one of these agents reaches a page that requires a login, "the agent app prompts you for the credential it needs from 1Password to continue." The user reviews that specific request and approves or denies it inside 1Password. If approved, 1Password auto-fills the credentials directly into the destination site — the agent is not handed the password, the one-time code, or any other vault contents it didn't request.

1Password frames this as a way to let agents do useful, logged-in work — booking something, filling out a form, checking an account — without the user having to hand over account access wholesale, the way copying a password into a chat window would. The company's blog post puts it plainly: integrations "design the integration to keep your credentials separate from the agent, so supported agents can continue working without needing to see or handle your credentials."

How the approval flow works

The mechanics described in 1Password's announcement follow a consistent pattern across partners, even though the exact controls differ by integration:

• An AI agent determines it needs to sign into a specific site to complete a task the user asked for.
• The agent's own app or interface prompts the user, naming the credential it wants to use.
• The user reviews that request and approves it inside 1Password — 1Password says users can see what will be shared before approving.
• Once approved, 1Password — not the agent or the AI model behind it — fills in the username, password, and any one-time code on the destination site.
• The user can disconnect any agent integration at any time.

This is the same basic pattern 1Password used when it first opened this door for AI coding and chat agents like Claude back in July: the model never sees the secret, only the outcome of being logged in.

Which agents are supported now, and which are coming

1Password's own announcement draws a clear line between integrations that are live at launch and ones still in development. As of the October 22 post, three agents have integrations available now:

• Muse (Meta) — available on iOS, Android, and Web.
• Grok Bot (SpaceXSI) — available on iOS, Android, Windows, macOS, and Linux.
• Townie (Town) — available on iOS, Android, Windows, Mac, and Linux.

1Password lists several more integrations as "coming soon," including Anthropic's Claude Duet and OpenAI's ChatGPT Work and dots, along with Instinct, Browserbase, Browser Use, Dana, KERNEL, Perplexity, Ramp, Tomo, and Wajo. Because 1Password itself describes these as not yet live, Pandromeda is not characterizing any of them as currently working integrations — only as agents 1Password says it plans to support. The company adds a standing invitation for feedback: "Don't see your favorite agent listed? We're building more every day."

1Password's post walks through what each live integration is meant to unlock. For Muse, the company's example is a party: a user can "ask an agent to create a party invitation, track RSVPs, and search for supplies on your budget," with Muse requesting access to sites like RSVPify and Oriental Trading as needed, while the user picks which 1Password login it's allowed to use. For Grok Bot, 1Password describes "AI teammates that work across tools, inboxes, and websites," using approved logins to "automate routine work more safely." For Townie, the pitch is a personal assistant that "learns their people, projects, and voice from their email, calendar, and work tools" and drafts, schedules, and follows up on the user's behalf — with the user approving whatever logins the Townie needs along the way.

Cost, eligible accounts, and regional limits

1Password says the new platform is "currently available at no additional cost for 1Password Individual and Family accounts" hosted on 1Password.com. The company is explicit that this is a regional and plan-based rollout, not a universal one: "Accounts on 1Password.eu or 1Password.ca aren't currently supported." 1Password's announcement does not list Teams or Business plans as part of this initial rollout. The company also notes that "availability, platforms, and capabilities vary by partner integration" — so even among supported agents, which platforms (web, mobile, desktop) an integration reaches differs case by case, as shown in the per-agent platform lists above.

Building on 1Password's earlier work with Claude

1Password for Agents is not the company's first attempt at letting an AI assistant use saved logins safely. In July 2026, 1Password and Anthropic introduced a browser integration that let Claude sign into websites using a saved 1Password login without the password ever being exposed to the model. That earlier integration established the core pattern the company has now generalized into a platform: the user approves a specific credential request, and 1Password — rather than the AI — is the one that actually types the secret into the page.

October's launch folds that same idea into a broader program with multiple AI partners at once, and lists Anthropic's newer "Claude Duet" agent as one of the integrations 1Password says is coming to the platform, rather than something it is building from scratch.

The July integration is worth understanding on its own terms, because it's the clearest publicly documented version of the mechanism 1Password is now generalizing. According to that announcement, when Claude needs to sign into a site, 1Password shows the user which credential is being requested and why; once the user approves with a biometric prompt such as Touch ID, "1Password injects the credential directly into the page," and, in the company's words, "Claude never sees the vault item, password, or one-time code." Access is scoped to the task at hand and ends when the task is done — under what 1Password calls Agentic Mode, the browser extension locks down once a compatible agent takes control, limiting it to only the logins approved for that specific task. 1Password's CTO, Nancy Wang, framed the reasoning behind the design this way: "The answer isn't handing agents your secrets. It is to let a user give an agent permission to use a credential without letting the agent see it. Claude knows it used your login; it does not need the password or one-time code in its context." That integration launched for the 1Password desktop app on Mac, spanning business, family, and individual plans — a wider plan range than October's agent platform, which so far is limited to Individual and Family accounts.

Why this matters for password security

The core problem 1Password for Agents is trying to solve is a familiar one in the password-manager world, just applied to a new class of software. As AI agents take on more tasks that require being logged into an account — filling out forms, comparing prices, managing bookings — the obvious shortcut is to just give the agent the password. 1Password's pitch is that doing so creates a durable secret the agent (and, by extension, whatever company trains or hosts its underlying model) could mishandle, leak, or reuse in ways the user never approved.

By keeping the credential inside 1Password's own vault and only ever "handing over" an authenticated session rather than a password, the company is applying the same reasoning that underpins passkeys and other credential-isolation techniques used across the password-management industry — a space Pandromeda has covered in comparing 1Password against rivals such as Bitwarden. It's also a response to the broader rise of agentic automation tools — the same trend behind the growth of workflow platforms like n8n, Zapier, and Make — which increasingly need some way to authenticate on a user's behalf without becoming a standing security liability.

1Password is careful not to oversell the guarantee, though. Its own language — "the safeguards and controls vary by integration" — signals that the strength of the protection depends on how each individual AI partner has implemented its side of the integration, not on a single, uniform standard 1Password enforces across the board.

Limitations and open questions

1Password's announcement focuses specifically on credential exposure — keeping the password and one-time code away from the agent and the model behind it. It does not claim to control what an agent does once it's signed in, such as what data it reads, what actions it takes inside an account, or how long a session stays active. Users approving a credential request are approving access to that login, not auditing everything the agent subsequently does with it.

It's also worth noting that most of the agents 1Password highlighted at launch — Claude Duet, ChatGPT Work and dots, and a long list of other named partners — are marked "coming soon," not live. Exactly when each of those integrations ships, and what controls they'll offer once they do, is something only 1Password and each partner can confirm as rollout continues.

What's next, and what this means for you

For now, 1Password for Agents is a free addition for 1Password Individual and Family subscribers on 1Password.com who use one of the three agents live at launch — Muse, Grok Bot, or Townie. If you're on 1Password.eu, 1Password.ca, or a Teams/Business plan, the feature isn't available yet, and 1Password hasn't given a timeline for when — or whether — that will change.

If you use one of the agents 1Password lists as "coming soon," including Claude Duet or ChatGPT Work and dots, there's nothing to set up yet; 1Password says it is still building those integrations. The more durable takeaway is the model itself: rather than typing a password into an AI chat window or an agent's own settings, the direction 1Password is pushing toward is approval-based, per-request credential sharing, where the password itself never has to leave the vault at all.

Written byFelix Moreau

Felix Moreau writes Pandromeda's software coverage and how-to guides. He covers Windows, macOS and Linux updates, the apps people rely on, emulators and developer tools, and turns official documentation into clear, numbered steps that work on the current version.

More from Felix MoreauAll Software & Guides coverage