Reasoning models struggle to control their chains of thought, and that’s good

OpenAI introduces CoT-Control, revealing that reasoning models struggle to fully control their chains of thought. This finding reinforces the critical importance of monitorability as an AI safety safe

• •OpenAI introduced a new method called CoT-Control to investigate reasoning model behavior.
• •The research found that reasoning models have difficulty fully controlling their internal chains of thought.
• •This discovery highlights and reinforces the crucial role of monitorability as an AI safety mechanism.