Huh? You don't mention the built-in sandboxing options in things like Codex. Why do people pretend these features don't exist?

https://learn.chatgpt.com/docs/permissions