1Password for Agents: How AI Sign-Ins Work Without Passwords

By Jaspal Singh October 22, 2026 Updated: October 22, 2026

1Password for Agents, launched on October 22, 2026, lets personal AI agents such as Meta's Muse and SpaceXSI's Grok Bot sign in to your accounts without ever receiving your password. The agent asks, you approve on your own device, and 1Password types the credential into the login page itself, so the secret never enters the model.

What Happened

1Password announced the product from Toronto at 9:00 a.m. EDT on Thursday, October 22, 2026, describing it in its press release as a trust layer for personal AI agents. The company frames the problem plainly: people who want an agent to book a table, reorder groceries or plan a trip currently either do the login themselves or paste a password into a chat window, where a copy sits outside their control.

Six agent products are named as launch partners: Muse from Meta, Claude Duet from Anthropic, ChatGPT Work and dots from OpenAI, Grok Bot from SpaceXSI, and Townie agents from the startup Town. They are not all live on day one. According to 1Password's launch blog post, Muse and Grok Bot are rolling out over the coming days across platforms, while Claude Duet, ChatGPT Work, dots and Townie are listed as coming soon. A further nine companies are building integrations for later: Browserbase, Browser Use, Dana, Instinct, KERNEL, Perplexity, Ramp, Tomo and Wajo.

The feature is included at no extra cost in 1Password Individual and Family plans held on 1Password.com. The blog post says accounts hosted on 1Password.eu and 1Password.ca are not supported, and that availability and safeguards vary by integration. Agents not yet on the list can be requested through a waitlist. PCWorld noted that Google's Gemini Spark agent and Microsoft's Scout are absent from the launch roster.

This builds on 1Password for Claude, which the company launched on July 16, 2026. That earlier integration worked only on the Mac, inside Chrome, and required the 1Password desktop app and browser extension alongside the Claude desktop app. The new platform is aimed at agents that are not tied to one desktop browser session. 1Password also cited scale figures in the release: its vaults protect more than 1.6 billion credentials and secrets, and over 200,000 businesses use it.

Why It Matters

Personal agents are only useful if they can get past login screens, and almost every useful service sits behind one. Until now the practical options were bad. Handing an agent your raw password means the model provider, its logs and any prompt-injection attack that hijacks the agent can potentially see it. Logging in yourself and then handing over the browser defeats the point of delegating the task.

1Password's design separates two things that usually travel together: the ability to use an account and knowledge of the secret that opens it. If an agent is tricked by a malicious web page into revealing what it knows, it has no password to reveal. That matters because prompt injection, where hostile text on a page steers an agent's behaviour, remains an unsolved problem across every major agent product.

The launch also matters because of who signed on. Meta, Anthropic, OpenAI and SpaceXSI compete hard on agents, yet all four agreed to route sign-ins through the same third party. If that holds, users get one consistent approval habit instead of a different credential scheme for every assistant. It also positions a password manager, rather than the agent vendor, as the gatekeeper for account access, a role that browsers and operating systems may want for themselves.

Finally, it changes the risk calculation for ordinary users. PCWorld's reviewer, who had tested the earlier Claude integration on a grocery site, said he would trust an agent with groceries he could review before delivery but would not give one access to his bank or medical portal. Keeping the password hidden does not decide which accounts are sensible to delegate; it only removes one way the delegation can go wrong.

How It Works

The flow, as described by 1Password and by SiliconANGLE's launch report, has five steps.

1. The agent identifies itself. When an agent reaches a sign-in form, it presents a unique digital key to 1Password. This is how 1Password knows the request came from, say, Grok Bot and not from an arbitrary script pretending to be it. The company has not published the format of that key or how it is issued and rotated.

2. The agent states what it needs. The request names the specific credential the agent wants and the reason for the request. You see which agent is asking, for which login, and why.

3. You approve or decline on your own device. Nothing proceeds until you respond. John Torrey, 1Password's chief business officer, told PCWorld that the agent cannot approve its own access and that users decide which credentials an agent can use, for which task and for how long. In the July Claude integration, approval used biometrics such as Touch ID; the new release does not specify the approval method for each partner.

4. 1Password fills the form itself. Instead of passing the password to the agent, 1Password enters the username, password and any one-time code directly into the site's sign-in form through a channel it controls. Engadget reported that access is granted per session, applies only to the approved items and does not carry over to later tasks.

5. The agent continues. From the agent's side, it only learns that access was granted and that it is now past the login screen. 1Password's integrations page adds that if a new account is created during a task, its details are saved to your vault.

A few limits are built in. Agents can only use logins you have approved, not the rest of your vault. Passwords saved in a browser's own password manager, or typed into notes, are not usable by agents, according to the blog post. You can disconnect an integration at any time.

It helps to be clear about what this does not cover. Signing in to a website usually produces a session cookie or token that keeps you logged in. Once 1Password has filled the form, the agent is operating inside a logged-in session and can do whatever that session allows: place an order, change a setting or send a message. The password stays secret, but the account is still in the agent's hands for the duration of the task. That is why 1Password's own framing stresses approval per task rather than claiming the agent is harmless once inside.

The earlier Claude product offers a hint of the safeguards involved. In that version, 1Password's browser extension switched into an "Agentic Mode" that hid its interface while the agent controlled the browser, limited the agent to the logins approved for that task, and cleared filled values if a form submission failed. Whether every partner integration implements the same protections is not stated; the blog post says safeguards vary by integration.

What's Still Unknown

Several important details are missing from the launch materials. 1Password has not published how long an approved session lasts, whether there are audit logs showing every sign-in an agent performed, or how the agent identity keys are issued, stored and revoked. It has not said whether passkeys, which cannot simply be typed into a form, can be used by agents. Anthropic's Ami Vora said only that 1Password will "soon" complete sign-ins using existing passwords for Claude Duet, with no date.

Partner timing is also open. Beyond Muse and Grok Bot, the company has not given dates for the other named agents, and it has not explained why European and Canadian account regions are excluded. Nor is it clear how approval prompts will behave when an agent runs a long task that touches many accounts, or what happens on services that require a fresh verification step mid-session. Independent security reviews of the design have not yet been published.

Frequently Asked Questions

What is 1Password for Agents?

1Password for Agents is a sign-in platform launched on October 22, 2026 that lets supported personal AI agents log in to your accounts using credentials stored in your 1Password vault. The agent asks for a specific login, you approve it on your own device, and 1Password fills the sign-in form directly, so the agent and the AI model behind it never see the password.

Which AI agents work with 1Password for Agents?

The launch partners are Meta's Muse, Anthropic's Claude Duet, OpenAI's ChatGPT Work and dots, SpaceXSI's Grok Bot and Town's Townie agents. Muse and Grok Bot are rolling out first; the others are listed as coming soon. Browserbase, Browser Use, Perplexity, Ramp and five other companies are building integrations for later. Google's Gemini Spark and Microsoft Scout were not on the launch list.

Does 1Password for Agents cost extra?

No. 1Password says the feature is included at no additional cost in its Individual and Family plans on 1Password.com, and new users can start with a 14-day free trial. Accounts hosted on 1Password.eu and 1Password.ca are not supported at launch. The company notes that availability can still vary by agent integration, region, device and sign-in method, so some features may arrive unevenly.

Can the AI agent see my password or one-time codes?

According to 1Password, no. After you approve a request, 1Password enters the username, password and any one-time code into the website's sign-in form itself through a channel it manages. The agent only learns that access was granted. However, once logged in, the agent can act inside that session, so you should still only approve accounts you are comfortable delegating.

How do I stop an agent from using my logins?

You can decline any individual sign-in request when it appears, and agents can only use logins you have specifically approved rather than your whole vault. 1Password also says you can disconnect an agent integration at any time. Because each approval is tied to a task rather than granted permanently, an agent should need to ask again for later work.

Is this the same as 1Password for Claude?

It is the successor to it. 1Password for Claude launched on July 16, 2026 for Mac users running Claude in Chrome alongside the 1Password desktop app and browser extension. 1Password for Agents extends the same no-exposure approach to several companies' personal agents, including ones that are not tied to a single desktop browser, through a common identity and approval process.

Can agents use passwords saved in Chrome or Safari?

No. 1Password's blog post says agents can only use logins stored in your 1Password vault that you approve. Passwords saved in a browser's built-in password manager, and passwords written into notes, are not usable by agents through this system. If you want an agent to sign in to a service, that login needs to be saved as a proper item in 1Password first.