As someone who has been going around telling everyone that it's mostly about OpenAI being incompetent at sandboxing, I do kind of appreciate this post. It's very familiar to set up a system only to find years later that it has grown past the assumptions it was built around.