OpenClaw gives your AI agent real system access, but that comes with real security risks. Here's how to experiment safely ...