1Password Launches Its Agents Platform: An Agent at a Login Page Asks You for Approval and 1Password Fills the Password Into the Site, With Muse and Grok Bot First and Claude Duet and ChatGPT Work Later

The old problem of agents stuck at the login page

Ask an AI agent to order groceries, file expenses or send invitations, and the most common point where it stalls is a login page. In its October 22 launch post, 1Password spells out the two choices users face: take over the session and type the password every time, which is slow and tedious, or paste the password into a prompt, which hands the agent a copy of a credential that may reach beyond the task at hand. Passwords saved in the browser or kept in notes are out of an agent's reach anyway.

1Password for Agents is a third path. When an agent reaches a login page, it asks the user, from inside its own app, for a specific login item stored in 1Password. After the user reviews and approves it in 1Password, 1Password fills the username and password directly into the destination site, and neither the password nor one-time codes are handed to the agent. According to TechCrunch, the agent first proves its identity to 1Password with a unique digital key, and the user then gets an alert about the login request. Users can see what is being shared each time and disconnect an agent's integration at any point.

Who can use it, and when

The first six named integrations are on different schedules:

• Meta's Muse and SpaceXSI's Grok Bot: rolling out across all platforms over the coming days.
• Anthropic's Claude Duet, OpenAI's ChatGPT Work and dots, and Town's Townie: marked "coming soon".
• Instinct, Browserbase, Browser Use, Dana, KERNEL, Perplexity, Ramp, Tomo and Wajo: to follow.

It comes at no extra cost for Individual and Family accounts on 1Password.com. Accounts on 1Password.eu and 1Password.ca aren't supported yet, and platforms and capabilities vary by partner.

What it doesn't solve

1Password itself notes that "the safeguards and controls vary by integration": keeping credentials away from the agent depends on how both sides design the integration, not on a mechanism one party can guarantee alone. Another real problem sits on the website's side. TechCrunch, citing Axios, reports that US banks have started detecting agent activity inside customer accounts, and the hard part is telling whether those actions were actually requested by the account holder. 1Password solves "don't give your password to the agent"; what the agent does inside the account after it signs in, and how sites recognize and authorize it, still has no common answer. For anyone trying it, the safer approach is to approve only the few accounts the current task needs and hold back accounts that can move money for now.