OpenClaw: The Most Dangerous AI Project on GitHub?

OpenClaw is a self-hosted AI agent with 200,000+ GitHub stars that connects to messaging apps, file systems, and terminals to act autonomously. Its architecture uses four layers: a WebSocket gateway, an LLM reasoning layer, a markdown-based memory system with write-ahead logging, and a skills execution layer. However, serious security issues have emerged: a WebSocket origin validation vulnerability allowed one-click full compromise, 20% of its plugin marketplace (Claw Hub) was found to contain malware, and over 30,000 instances are exposed on the public internet with no authentication. Six additional CVEs were recently disclosed. Safe usage recommendations include running it in Docker or rootless Podman with two-layer container isolation, binding the gateway to localhost only, and vetting every plugin before installation.