1Password for Agents, which 1Password launched on October 22, 2026, lets a personal AI agent sign in to a user’s accounts without the agent ever receiving the password. The agent asks 1Password for a login, the person approves the request on their own device, and 1Password fills the credential straight into the website’s sign-in form. What the launch protects is the password itself. The account session the agent works in after that approval is governed by each agent maker’s integration, and 1Password says those safeguards vary.
1Password for Agents Fills the Login So the Agent Never Holds It
The company’s announcement, issued from Toronto, describes a three-party exchange. When an agent reaches a sign-in form, it presents a unique digital key to 1Password to prove which agent it is. The person is shown which credential the agent wants and the reason for the request, and decides on their own device whether to allow it. 1Password then types the credential into the form instead of passing it to the agent, and the agent carries on with its task. In a launch blog post, 1Password chief business officer John Torrey writes that one-time codes stay hidden from the agent as well. Chief product officer Aashish Tripathi framed the problem as people who “hand over a credential and hope for the best.”
Pasting a Password Into a Prompt Leaves Copies Nobody Controls
1Password’s case rests on what people do today when an agent stops at a login page. Torrey’s post lists two options: take over the session and type the password by hand each time, or paste it into the prompt and give the agent the credential for longer than the task needs. The press release adds that a credential handed to an agent can end up stored for next time in places the owner can no longer manage. The new platform replaces both habits with a request the person sees and approves. Because the credential stays in one vault, a changed password updates once and every connected agent picks up the new one.
Most of the Launch Integrations Are Not Live Yet
1Password named five agent makers at launch: Meta’s Muse, Anthropic’s Claude Duet, OpenAI’s ChatGPT Work and dots, SpaceXSI’s Grok Bot, and Town’s Townie assistant. The blog post gives a status for each. Muse and Grok Bot are rolling out over the coming days, while Claude Duet, ChatGPT Work, dots and Townie are listed as coming soon. Nine more companies are building integrations, including Browserbase, Browser Use, Perplexity and Ramp. The feature costs nothing extra on Individual and Family plans hosted on 1Password.com, and accounts on 1Password.eu and 1Password.ca are not yet supported. Anthropic chief product officer Ami Vora said in the release that Claude Duet already asks before it signs in anywhere.
The Agent’s Approval Ends at the Sign-In Form
The design keeps one secret out of the agent’s reach and leaves the rest of the job to the agent. Once 1Password completes the login, the release says the agent continues its task inside the account. Neither the release nor the blog post describes per-action limits, session lengths or an activity log for what an agent does after it is signed in. 1Password states that it and each partner design the integration together, and that the safeguards and controls differ from one integration to the next. TechCrunch’s Lucas Ropek noted that the launch landed on the same day Axios reported that US banks are detecting agent activity inside customer accounts and often cannot tell whether the customer asked for it.
Why This Matters for Engineering Leadership
Engineering teams face the same choice 1Password describes, with API keys and service tokens in place of household logins. An agent that needs to reach a cloud console or a SaaS tool is usually given the token in its configuration or its prompt, where it can be logged, cached or copied. 1Password’s model is a pattern worth requiring of internal agent tooling: a broker holds the secret, the agent proves its own identity, a human approves the specific request, and the secret is injected where it is used. The launch also shows where that pattern stops. Keeping the password out of the model does nothing to bound what the agent does with the access it gains. Before approving any agent integration, ask the vendor three things. Find out what the agent holds after sign-in, how long that access lasts, and where its actions are recorded. For a retailer’s answer to the same problem, see our report on Walmart’s AI agent profiles.
FAQ
What is 1Password for Agents?
It is a 1Password platform launched on October 22, 2026 that lets personal AI agents sign in to a user’s accounts without receiving the password. The person approves each request and 1Password fills the credential into the sign-in form.
How does an AI agent sign in with 1Password for Agents?
The agent verifies its identity to 1Password with a unique digital key, and the person sees which credential it wants and why. After the person approves on their own device, 1Password fills the login directly into the site.
Which AI agents work with 1Password for Agents?
Launch integrations cover Meta’s Muse, Anthropic’s Claude Duet, OpenAI’s ChatGPT Work and dots, SpaceXSI’s Grok Bot and Town’s Townie. 1Password says Muse and Grok Bot roll out over the coming days and the others are coming soon.
Does the agent ever see the password or one-time code?
1Password says neither the password nor the one-time code is exposed to the agent. Once signed in, the agent continues working inside the account.
Who can use 1Password for Agents and what does it cost?
It is available at no additional cost on 1Password Individual and Family accounts hosted on 1Password.com. Accounts on 1Password.eu and 1Password.ca are not supported yet.