What 1Password launched

On October 22, 2026, 1Password released 1Password for Agents, a way for personal AI agents to sign in to your accounts without ever being handed the password. The agent asks for a login, you approve it on your own device, and 1Password types the credentials into the site itself. The agent is told one thing: access granted.

The launch partners are the companies now selling agents to consumers. Meta's Muse and SpaceXSI's Grok Bot are first, with Anthropic's Claude Duet, OpenAI's ChatGPT Work and dots, and Town's Townie listed as coming soon. For anyone with an Individual or Family plan on 1Password.com, it costs nothing extra.

The problem it is aimed at

Agents that book, buy and file things for you spend much of their time in front of sign-in pages, and that is where they stall. 1Password's launch post describes the two options people have had so far: take over the agent's session and type the password yourself every time, or paste the password into the prompt and let the agent hold it, "potentially beyond the task at hand." The company's verdict on both: "Neither of those are acceptable."

The second habit is the one that worries banks. TechCrunch reports that U.S. banks have already detected agent activity in customer accounts, and a Chase spokesperson repeated the old rule that customers "should not give their usernames and passwords to anyone." A password pasted into a chat sits in a conversation log you no longer control.

There is a quieter failure too. People who keep passwords in the browser or in a notes file have found their agents cannot use them at all, so every login means stopping the agent and signing in by hand.

How a login works now

1Password's product page breaks it into three steps. At the login screen, the agent asks for the credential for that account. You approve or deny the request in 1Password, and any new account details are saved to your vault. Then 1Password fills in the password or one-time code, and the agent "only sees that access is granted."

According to TechCrunch, the agent proves who it is with a unique digital key when it reaches the form, and the person sees which credential it wants and why before approving. Because approval happens on your own device, 1Password can confirm the yes came from you rather than from the agent.

The scope is per request, not per vault. Connecting an agent does not open everything you have stored: it asks for one login for one task. If several saved logins match a site, you choose which one gets used. You can also disconnect an agent at any time, and because the password lives in one place, changing it once keeps every connected agent working.

1Password's own example: ask Muse to plan a party, track RSVPs and shop for supplies on a budget. Muse asks for access to RSVPify and Oriental Trading as it needs them, you pick the login each time, and it never reads the credentials or touches anything else in the vault. Other examples in the post are turning a grocery list into an Instacart order and trip receipts into expense reports.

Who can use it, and where

As of October 24, 2026, here is where each named agent stands, per 1Password's launch post. TechCrunch and SiliconANGLE describe OpenAI and Town as working at launch; 1Password's own post says coming soon, and that is what this table follows.

It is included in every Individual and Family plan, and new users get the standard 14-day trial. Two exclusions matter. Accounts hosted on 1Password.eu or 1Password.ca are not supported yet, so many European and Canadian customers will have to wait. And 1Password names only Individual and Family plans, saying nothing about Business or Teams accounts. There is no waitlist for the feature itself; the waitlist on the product page is for asking 1Password to add an agent it does not list.

What it changes

For people using agents, the change is a habit to drop. Pasting a password into a chat stops being the quick path, because there is now one that is just as fast. It only helps if your passwords already live in 1Password, though. Anyone still relying on the browser's built-in manager has a migration to do first. Our 1Password review covers the plans and what they cost.

For people building agents, 1Password is offering to own the sign-in step: "You build the agent, 1Password secures the sign-in." That means not storing user credentials, not writing a vault, and not answering for a leaked password. What it does not offer yet is anything public. No SDK, API reference or builder pricing was published at launch; developers are pointed to 1Password's partnerships team. A small team cannot plug in this week the way Meta and SpaceXSI have.

The partner list is the other story. Meta, Anthropic and OpenAI all lent quotes to the press release. "The old model where you hand over a credential and hope for the best doesn't work," said Aashish Tripathi, 1Password's chief product officer. Anthropic's chief product officer Ami Vora put it from the agent's side: "Most of the work people want help with lives inside their accounts, and we know an agent has to earn its place there."

What it does not solve

The agent never sees the password, but it does get the logged-in session. Once you approve a login to your bank or your Amazon account, the agent can do whatever that session allows. 1Password controls the key to the door, not what happens inside.

How much protection you get also depends on who built the agent. 1Password says so plainly: "The safeguards and controls vary by integration," and availability and capabilities "vary by partner integration." The same approval flow can sit on top of a careful agent or a careless one, and the launch material does not publish a minimum standard partners must meet.

What comes next

Muse and Grok Bot users should see the integration over the coming days. Claude Duet, ChatGPT Work, dots, Townie and the nine smaller partners have no dates. Per SiliconANGLE, this builds on 1Password's earlier agent work, including Secure Agentic Autofill with Browserbase in late 2025 and browser integrations with Perplexity and Claude. The open questions are whether a public developer path follows, and when EU and Canadian accounts get support.