Claude's Computer use is great but security risks involved is terrifying.

Anthropic has released a research preview of Claude Computer Use for macOS, giving its AI model direct access to users' screens to assist with development tasks. The feature captures constant screenshots of active windows, allowing Claude to debug React components, fix CSS issues, and read browser console errors in real-time—functioning like a tireless senior pair-programmer. This capability effectively makes expensive $600 Mac Mini setups for OpenClaw obsolete, offering similar functionality through a $20/month Claude subscription directly on developers' machines.

However, the security implications are significant and concerning. The feature creates multiple attack vectors, most notably through prompt injection where malicious websites could gain indirect access to users' local environments. China has already issued strict warnings and bans on similar OpenClaw technology for government use due to these exact risks. Enterprise adoption faces major barriers as no serious organization would allow such permissions on bare metal without mandatory containerization in Docker or VM sandboxes. While technically proficient developers can mitigate risks, average users face potential data breaches from AI hallucinations or compromised instructions.