Except that, according TFA, even OpenAI obscured or didn't even notice some of the worst implications of what the agents surreptitiously did.