Chrome Could Let Gemini Agents Access Saved Passwords With User Approval
Google is building an experimental Chrome system that could let Gemini agents request access to selected saved passwords after users authenticate the request.
Recently, we reported that Gemini can help you change compromised passwords in Chrome. Google is now working on a new Chrome feature that could let Gemini agents access selected saved passwords when they need to sign in to websites.
The feature is still experimental, but Chromium code spotted by Windows Report shows that Google is building a dedicated password-sharing system for remote agents. A new #remote-actor-credential-sharing flag describes the feature as allowing “remote actors to request passwords from the signed in Google account.”

Chrome would not simply hand over a saved password, however. The current implementation is designed to ask the user to select a password and authenticate before the credential can be shared.
The code refers to granting permission to a Gemini Agent as part of the password-sharing process. The Chromium commit describing the sharing flow explicitly says it will upload the credential to Passbox and then “grant permission to the Gemini Agent in APS.”
The shared credential can have a limited lifetime, while the permission can also be revoked. The design points to temporary access rather than unrestricted access to a user’s saved passwords.
Chrome is also adding a private chrome.requestAgentAuthentication API for approved origins. The API appears to provide an entry point for requesting agent authentication, although the current implementation still includes a no-op selection path.
The implementation includes the credential-selection dialog and the components needed for much of the credential-sharing flow.
Development notes also say the end-to-end backend flow was manually verified against a staging environment, with successful responses from Passbox and the Agentic Permission Service. This was an engineering test rather than a public Chrome feature test.
For users, the bigger implication is that Gemini could eventually handle tasks that require signing in to websites on their behalf. Instead of asking users to manually enter a password during an agentic task, Chrome could provide a controlled way for the agent to request access to a specific saved credential.
This does not mean Gemini can access Chrome’s saved passwords today. The feature is still under development and is not exposed as a normal Chrome feature.
The work provides an early look at where Google’s agentic browsing plans could be heading: Chrome could let Gemini agents use saved credentials for specific tasks while keeping the user involved in password selection and authentication.
Read our disclosure page to find out how can you help Windows Report sustain the editorial team. Read more
User forum
0 messages