I think people are kind of missing the ”problem” with @clawdbot - its not the security model. That can be managed with separate accounts and a clean machine. The real issue is privacy - it can be incredibly useful IF you give openai or anthropic free access to all your data.