I see OpenAI are giving the exact same bullshit defense in his case too. “We’ve updated the model so now it won’t libel you specifically, but it will keep libelling just about anyone else if the prompt isn’t worded in exactly the right way”