1Password Launches Agentic Authentication Feature for Claude

A feature has been launched that allows Anthropic's Claude to securely authenticate when logging into a user's account to perform tasks, without directly exposing passwords. Implemented through a collaboration between 1Password and Anthropic, this approach focuses on addressing security vulnerabilities in AI agents.

Key Mechanism and Features:

• No Password Exposure: When Claude reaches a login screen, 1Password asks the user which account to use, and once the user approves via biometrics (fingerprint/face), 1Password directly enters the password and OTP (one-time password). Throughout this process, Claude never sees the password or security code.
• Scoped Access: Access permissions are limited to that specific task and end immediately once the task is complete.
• Security Verification: After login, the page is checked for data leaks, and if login fails, the entered information is immediately deleted to maintain security.

This feature is currently available to Personal, Business, and Family plan users using the 1Password desktop app for Mac and browser extension. It is regarded as core security infrastructure that enables AI agents to move beyond simple demos and perform actual work.