Latest / Frontier models and safety
OpenAI studies what changes how well chain-of-thought can be monitored
See how this story connectsOn 10 Nov 2026 OpenAI published a study of how pretraining, architecture, reinforcement learning and model-generated data affect two proxies for chain-of-thought (CoT) monitorability. It found that no-CoT capability rises smoothly as base-model perplexity falls, and that CoT controllability increases during reinforcement learning and can pass between models through supervised fine-tuning on model-generated data. The authors call these imperfect proxies and say that models will become harder to monitor through their CoT as they scale toward more intelligent systems.
Why it matters
Safety teams that rely on chain-of-thought to detect misalignment face models OpenAI says are becoming harder to monitor as they scale.
- When
- 10 Nov 2026
- ID
WTA-N-2026-0873- Where
- International
- Who
- OpenAI
Summary written with AI by wheresthe.ai, from OpenAI and the reporting credited above. Checked against the source on 12 Nov 2026.
Source copy: saved on 12 Nov 2026 to check this story and not republished, because the text belongs to its publisher. Document the source offers: document 1 (PDF). Also held by the Internet Archive.
Cite this story
WTA-N-2026-0873: “OpenAI studies what changes how well chain-of-thought can be monitored.” wheresthe.ai, 10 Nov 2026. https://wheresthe.ai/d/intl-openai-studies-what-changes-how-well-chain-of-thought-c-2026-11/ (source: OpenAI).
Spotted an error? Email [email protected].