Anthropic's Claude Cowork and Claude Code can now take control of a Mac's mouse and keyboard, acting as another user to perform tasks like sending files or starting development servers. This feature, detailed in AppleInsider, allows the AI to interact with locally stored files and applications directly, representing a significant step beyond typical chatbot queries. However, the company explicitly warns this represents an early-stage security risk, advising users to only grant access to trusted apps handling non-sensitive data due to potential vulnerabilities.

The move positions Anthropic to compete with rivals like OpenClaw and Perplexity's Personal Computer, though Anthropic emphasizes caution over unrestricted access.