1Password lets Claude log into websites without seeing your passwords

Password manager 1Password has introduced an integration designed for Anthropic’s AI model Claude, alongside a broader browser feature dubbed Agentic Mode. The setup lets an autonomous browser agent complete tasks behind authenticated portals without exposing plaintext credentials to the model itself.

When Claude encounters an authentication wall, 1Password intercepts the action and shows the user an explicit prompt explaining which account is being accessed and why. Once the human confirms the request using biometric verification, the extension directly populates the login form and one-time two-factor authentication (MFA) codes on the webpage. The AI never receives the vault entry, master password, or temporary code in its working context window.

According to 1Password Chief Technology Officer Nancy Wang, access remains strictly scoped to the active task. Once the agent completes its workflow, access expires. The extension also inspects the page to confirm that credentials do not linger in visible form fields, and wipes any inserted values if the login submission fails.

The company is positioning the underlying safeguard, Agentic Mode, as a universal browser defense for third-party automation. When a compatible AI agent takes control of the browser, 1Password locks its extension interface, rendering the rest of the vault inaccessible beyond the explicitly authorized credentials.

Separately, rival privacy suite Proton announced that its password manager now supports storing software licenses, Wi-Fi credentials, and SSH keys, with a desktop biometric SSH agent and an offline mode currently in development.

Source: habr

Primary announcement: 1Password