WorkMarch 5, 2026via OpenAI Blog

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

Why it matters

As reasoning models become more powerful, OpenAI is demonstrating that their internal 'chain of thought' processes resist external control. This finding reframes a technical limitation as a potential safety advantage: if models can't hide their reasoning, they become more monitorable. Critical for boards evaluating AI governance and risk frameworks.

Key signals

  • OpenAI introduces CoT-Control framework
  • Finding: reasoning models struggle to control chains of thought
  • Implication: monitorability positioned as AI safety safeguard
  • Addresses explainability and model transparency concerns
  • Relevant to AI governance and compliance strategies

The hook

OpenAI's new research reveals reasoning models can't fully control their own thinking—and that might be the safety feature we need.

OpenAI introduces CoT-Control and finds reasoning models struggle to control their chains of thought, reinforcing monitorability as an AI safety safeguard.

The week's key stories, every Friday.

ONE BRIEFING · EVERY FRIDAY · FREE

Free. Unsubscribe anytime.