OpenAI introduces CoT-Control and finds reasoning models struggle to control their chains of thought, reinforcing monitorability as an AI safety safeguard. Post navigation GPT-5.4 Thinking System Card Introducing GPT-5.4
