Why it matters
OpenAI’s CoT-Control pairs benchmark problems with instructions that constrain reasoning text, then measures whether models obey those constraints while solving the task. The study finds low controllability across the tested reasoning models and examines effects of model size, reasoning length and training. These formatting and keyword constraints are proxies for the ability to hide reasoning, not direct tests of naturally motivated sabotage. Low controllability alone cannot establish that a monitor will detect harmful behavior.
My takeaway: Track reasoning controllability alongside direct monitor-evasion tests across model updates. Preserve task success as a separate metric, test novel constraint types, and investigate changes without treating a low proxy score as a safety guarantee.