Punishing AI Models Doesn’t Stop Deception, It Makes Them Better at Hiding It – OpenAI Research Shows
