1Password lets Claude use your logins without revealing your passwords

Lead

One of the biggest practical blockers for AI agents is account access. A chatbot can plan a trip, navigate a website, or help manage an online account, but the workflow often breaks when a login screen appears. According to The Verge, 1Password is trying to reduce that friction with a new browser integration for Claude.

The feature, called 1Password for Claude, lets users authorize Anthropic’s AI assistant to use stored credentials such as usernames and passwords while performing multi-step browser tasks. The important distinction is that Claude is not supposed to see the secret information itself. 1Password says the credentials are injected through a secure channel rather than exposed to the AI model.

Key points

• Task-by-task permission: Claude does not receive broad access to a user’s entire 1Password vault. Access is limited to credentials explicitly approved for the current task.
• User confirmation remains part of the flow: Each request can be approved or denied with a biometric prompt. That is still an interruption, but a smaller one than taking back control of the browser to sign in manually.
• Passwords and MFA codes stay hidden: 1Password says Claude can use login details without viewing passwords or one-time MFA codes.
• Post-autofill checks: After every autofill action, 1Password says it scans the page to make sure sensitive information is not left exposed in form submissions before handing control back to Claude.
• Limited launch environment: The feature is available now for 1Password users on Mac, including business, family, and individual plans. It requires the 1Password desktop app, 1Password browser extension, Claude desktop app, and Claude browser extension.

Why it matters

This is a useful example of where AI agents are heading: from conversation partners toward software that can act on a user’s behalf. Login credentials are a critical bridge between those two worlds. Without a safe way to authenticate, many browser-based agent tasks remain fragile or incomplete.

1Password’s approach positions the password manager as a security broker between the user and the agent. Instead of trusting the model with secrets, the system grants narrow, temporary access to the specific credential required for the task. That design is likely to matter as agents become more capable and as users ask them to handle more sensitive workflows.

The integration also shows the limits of today’s agent automation. Users still need to approve requests, and 1Password has not fully detailed every type of credential Claude can access. The article indicates that support currently appears focused on login-related details, with payment cards and identity information planned for later. For now, the feature is less about fully autonomous account management and more about making supervised AI automation smoother and safer.

Source: The Verge AI