Overview
- Anthropic announced Claude Duet as a waitlisted research preview on Oct. 21, opening access only to US Claude Pro and Max subscribers and excluding Team and Enterprise for now.
- The agent runs on Anthropic’s Opus 5.5 inside a dedicated cloud virtual machine with its own browser and links to more than 6,000 connectors from the Claude Marketplace so it can keep working when a device is off.
- Claude Duet is built to act proactively by delivering a daily briefing, advancing tasks across email, calendar and business apps, and requesting user approval before sending messages, making payments, signing documents, or deleting content.
- Anthropic added safety controls including a dedicated classifier that blocks sharing the strictest data such as credit card numbers, and it prohibits high-risk actions like trading securities or moving funds between accounts; purchases use Stripe Link or Shop Pay with virtual cards for privacy.
- The release places Duet in direct competition with OpenAI’s Dots, Meta’s Muse and xAI’s Grok Bot and raises practical concerns about visibility into agent actions, prompt‑injection risks, permission rules for business systems, and faster consumption of plan quotas as the preview matures.