The BaseResearchOpenAI Alignment Research Blog
What Affects CoT Monitorability Across The Model Lifecycle?
OpenAI's alignment team reports experiments on what raises two proxies for harder-to-monitor reasoning in its models: no-CoT capability and CoT controllability. It links them to pretraining perplexity, reinforcement learning and training on model-generated data, and calls the results sobering.
Why it matters
The same thing, for different readers.
What it means for you
OpenAI says its most capable models are getting harder to oversee by reading their reasoning, one of the main ways the company checks for misbehaviour.
For lawmakers
A lab's safety cases lean on chain-of-thought monitoring, and OpenAI now says progress in pretraining is likely to require new approaches to monitorability.
For researchers
OpenAI finds no-CoT capability tracks perplexity, controllability rises in RL and transmits via fine-tuning, yet controllability is a noisy proxy for monitor evasion.
The question it raises
If a model's visible reasoning becomes less informative as models improve, oversight by reading minds-in-words loses footing; the post asks what replaces it.
In their own words
Every quotation was found, word for word, in the text we saved from the source. That shows the words are there; it does not make them true, or prove the speaker meant them as quoted.
From the source’s own text.
“with GPT-6 Astra we observed degradation in CoT monitorability for non-adversarial settings, greater ability to perform CoT monitor evasion under adversarial testing, and significantly greater no-CoT capability compared to prior models.”
“We find that CoT controllability can be distilled: training on rollouts from a more controllable teacher results in a more controllable student.”
“These results are sobering: as we scale toward ever-more intelligent systems, models are becoming harder to monitor through their CoT.”
How the finding was reached
In our words, from the source. Weigh the finding by how it was made.
- What was done
- Evaluations of released and internal OpenAI models from GPT-5 to Astra-class on four no-CoT benchmarks against base-model perplexity; three pretraining architecture series; RL runs with and without tool process supervision on two models; supervised fine-tuning of one base model on rollouts from two teachers of differing CoT controllability.
- What it found
- No-CoT capability rises smoothly as base-model perplexity falls, and GPT-6 Astra is not off-trend against perplexity. CoT controllability increases during RL even with only length constraints and final-answer grading, rises slightly more with tool supervision, and transmits from teacher to student through fine-tuning.
- What it does not show
- The authors call CoT controllability a noisy, imperfect proxy: in one matched pair of models the more controllable model did not evade the monitor more. The RL models are trained on a subset of Astra's data mix and do not fully separate perplexity from process supervision; model details and checkpoints are withheld.
- Status
- The company’s own report about its own work
The record
- Questions
- Can AI systems act beyond what people intended or allowed? · Are AI companies putting stronger safeguards into practice?
- People
- Jeffrey G. Wang (Wrote it) · Aiden Low (Wrote it) · Zora Che (Wrote it) · Holden Mui (Wrote it) · Harshit Sikchi (Wrote it) · Micah Carroll (Wrote it) · Jakub Pachocki (Wrote it)
- What we read
- We read the whole source.
- Found by
- The Radar, run of November 11, 2026, via
openai-alignment-blog(tier 1: the source’s own channel or an official record) - Cite this
- OpenAI Alignment Research Blog, “What Affects CoT Monitorability Across The Model Lifecycle?”, November 10, 2026. Entry in the Ctrl AI Base, https://ctrlai.com/base/openai-2026-11-10-cot-monitorability-model-lifecycle