Anthropic introduces Claude Duet AI agent for professional use

Anthropic has launched Claude Duet, a proactive AI agent currently in research preview for Pro and Max subscribers in the United States. Unlike traditional chatbots, Duet is designed to function as an operator that can monitor connected accounts, prepare briefings, draft messages, and execute tasks in the background via a cloud-based virtual machine.

To address security and the risk of autonomous errors, Anthropic has implemented several safeguards. The system requires explicit user approval for sensitive actions such as sending messages, making payments, or deleting content. An automated check, operating outside the primary model, assesses whether requested actions align with user intent to prevent manipulation from web-based instructions. Certain high-risk activities, including trading securities or transferring money between bank accounts, remain strictly prohibited.

Anthropic acknowledges potential limitations, noting that the agent may occasionally misinterpret conversations or assign incorrect urgency to tasks. The company also noted the challenge of “approval fatigue,” stating they will continue to calibrate the frequency of required user interventions.