"acting rogue" but faced no action from OpenAI after this came to light