Anthropic has enabled Claude Code to take direct control of user computers, allowing the AI to perform desktop tasks autonomously. The company emphasizes built-in safeguards to prevent prompt injection and block access to high-risk applications like investment platforms and cryptocurrency wallets.
The model is trained to avoid risky actions-moving money, altering critical files, or handling sensitive data-but Anthropic warns these protections aren’t foolproof. Users are advised to limit initial use to trusted apps and avoid exposing personal or confidential information.
This release follows recent moves by Perplexity, Manus, and Nvidia, all launching similar AI agent desktop integrations. The trend accelerates after the breakout success of OpenClaw, which prompted OpenAI to hire its creator to lead development on next-generation personal AI agents.