Anthropic details safeguards keeping Claude Duet agent in check

Anthropic outlined the security and privacy protections behind Claude Duet, its proactive AI agent that runs in a cloud browser and completes tasks while users are away. Actions like sending sensitive messages, making payments or deleting data require approval, with a separate check outside the model verifying intent. The standalone app is in a research preview waitlist for Pro and Max subscribers in the US.

• Passwords, cookies and sign-in tokens stay outside the model and are entered via a form in the Claude app
• Payments and account creation require approval every time, even with prior consent
• A separate classifier scans outgoing messages and browser uploads for sensitive information
• An external penetration test with source code access and a public HackerOne bug bounty cover Duet