The 1Password AI agents feature lets compatible assistants sign in to websites using credentials stored in a user's vault without revealing passwords to the agent. Announced on October 22, the feature requires the user to approve each credential request before 1Password fills the login directly into the destination site.
When an agent reaches a sign-in page, it requests the credential it needs and explains why. The user can approve or reject the request on their own device. 1Password says it verifies that requests come from authorized partners whose integrations have been reviewed for security. The agent receives access to the login action, not the password itself.
Which AI agents are supported?
The rollout begins with Meta Muse and SpaceXSI's Grok Bot, with integrations arriving over the coming days. 1Password says support for Town's Townie, Anthropic's Claude Duet, OpenAI's ChatGPT Work and dots is coming soon. More integrations are also planned with partners including Browserbase, Browser Use, Instinct and Perplexity. Availability, supported platforms and capabilities can vary by integration.
The feature is available at no additional cost for 1Password Individual and Family accounts hosted on 1Password.com. Accounts on 1Password.eu and 1Password.ca are not currently supported. The rollout is opt-in, so connecting an agent to 1Password is a separate choice rather than automatic access to a user's vault.
Password protection does not remove every agent risk
Keeping passwords out of an AI model's context reduces one way credentials could be exposed, but it does not make every action an agent takes safe. Once signed in, an agent may still be able to view account data or perform actions allowed by that account. Users should review each request and consider the permissions available inside the destination service.
In its official announcement, 1Password describes the feature as an opt-in way to let agents request only the credentials needed for a task. The approach builds on its earlier work on 1Password for Claude, which introduced credential access for Claude. Readers assessing broader agent permissions can also review Saganote's coverage of macOS Full Disk Access and AI agents and protecting data when using AI chatbots.
Related Articles
Google's Selfie Video Sign-In Lets a Face Recording Replace Your Recovery Phone Number
ChatGPT iMessage Plugin Arrives on Mac to Read, Draft, and Send Messages
Perplexity Hybrid Compute Brings Private AI Work to Mac
Meta Muse vs Grok Bot vs OpenAI Dots: Which Personal AI Agent Fits Your Workflow?
Google Passkeys: How to Set One Up and Sign In Without a Password
OpenAI Killed Its Browser. Now ChatGPT Wants Control of Your Chrome Tabs.