1Password will let AI agents sign in for you, free, without ever seeing the password

1Password has launched 1Password for Agents, a platform that lets a personal AI agent sign in to a website after you approve the request, while 1Password fills in the credential so the agent never receives it. The company announced it on October 22, as TechCrunch reported.

Free for personal plans, live for only two agents

According to 1Password’s blog post, the platform is available “at no additional cost” to 1Password Individual and Family accounts on 1Password.com. Accounts on 1Password.eu and 1Password.ca are not supported. The company adds that availability, platforms and capabilities vary by partner integration.

The rollout is uneven. Meta’s Muse and the Grok Bot from SpaceXSI are described as arriving over the coming days across all platforms. Anthropic’s Claude Duet, OpenAI’s ChatGPT Work and dots are listed as coming soon. So are Town’s Townie (web and iOS), Instinct, Browserbase, Browser Use, Dana, Kernel, Perplexity, Ramp, Tomo and Wajo. People who do not see their agent can join a waitlist, and developers are asked to contact 1Password’s partnerships team.

The agent asks, you approve, 1Password types

The flow, as 1Password describes it, has four steps. The agent uses a unique digital key to prove its identity to 1Password. You see which credential it wants and why, then approve or decline on your own device. 1Password enters a password or one-time code directly into the sign-in form. The agent then continues its task.

The company’s example is a party-planning task in which Muse asks for access to RSVPify and Oriental Trading. The user picks which login Muse may use, and Muse cannot see the credentials or reach anything else in the vault. Users can see what will be shared and disconnect an integration at any time. Because the credential stays in one vault, a password change only has to be made once.

Aashish Tripathi, 1Password’s chief product officer, said in the company’s announcement: “The old model where you hand over a credential and hope for the best doesn’t work,”.

A bet that agent logins become a paid layer

The launch fits a pattern for 1Password. Earlier this year it released a browser-extension feature for Anthropic’s Claude that, according to Engadget, lets Claude use stored credentials while keeping private data away from the model. Its developer documentation also cautions against passing secrets to a model where possible and favors short-lived, tightly scoped tokens. The new platform widens that approach from one assistant to a roster of them.

The commercial context is also notable. Techmeme, summarizing a Wall Street Journal report, says 1Password has passed $500 million in annual recurring revenue, up from $400 million in November 2025, driven by enterprise AI coding assistants and agents. A free consumer feature does not add revenue directly, but it puts 1Password inside the sign-in step of every agent that integrates. Neither the company’s materials nor the reports say how it plans to monetize that position, so that remains open.

Other companies are working on the same trust problem from different directions. Our earlier coverage of ServiceNow’s Arc agent noted its revocation of access in about a second, an enterprise-side answer to controlling what an agent can do. 1Password is aiming at the consumer side, where the user is the administrator.

Banks are already seeing agents in accounts

TechCrunch notes the launch came on the same day Axios reported that U.S. banks have begun detecting agentic activity inside customer accounts, and that most cannot tell whether it reflects a user’s request. A Chase spokesperson told Axios, per TechCrunch, that customers “should not give their usernames and passwords to anyone.” That is the habit 1Password’s design tries to remove: the agent holds no copy that could be saved for next time.

What the announcement does not settle

The security claims are 1Password’s own and have not been independently tested. The company says safeguards and controls vary by integration, so protection will depend on how each agent maker implements the handoff. The approval prompt also puts the decision on the user, who must judge whether a request is legitimate. Most of the named agents are still marked coming soon, no dates were given for them, and the materials do not describe business or enterprise availability. For now, the practical change is limited to Individual and Family customers on 1Password.com who use Muse or Grok Bot.