I haven't looked into it myself, but didn't OpenAI go through another company for sandboxing? Wouldn't failure of containment place liability in their hands?