Anthropic launches Claude Duet, a proactive AI agent for background tasks

Anthropic has introduced a research preview of Claude Duet, an autonomous AI agent designed to handle tasks in the background, even when offline. According to the announcement, the app relies on the Claude Opus 5.5 model and connects to over 6,000 services through dedicated connectors, including Gmail, Slack, and Stripe.

The agent sends a daily summary of priority items from connected apps and only intervenes when necessary. It operates on a dedicated virtual machine, avoiding strain on the user's device. Tasks are assigned via text or voice message in a single thread, covering actions like scheduling, financial forecasting, or prototype creation. Before any sensitive action (payment, message sending, document signing), the agent requests explicit approval.

This move follows a wave of similar products launched recently by xAI, Meta, and OpenAI. Anthropic emphasizes that Duet cannot trade securities or move funds between accounts, even with approval. The company highlights technical safeguards, such as a sensitive data filtering system and external penetration tests to mitigate risks of leaks or prompt injection attacks.