How OpenAI’s Control Room Can Cut Off a Misbehaving AI Model in 35 Seconds - Blockonomi