CoTs remain monitorable when being unfaithful requires computation: Replicating and Extending the Necessity Effect Across Eleven Models

Key points

• While a LLM's chain of thought (CoT) is a promising approach to interpreting model outputs, prior work shows CoT's content does not always accurately reflect the reasoning process.1
• We replicate their proxy task across eleven models from six families and find the necessity effect generalizes.1
• The relationship between when CoT is necessary and unfaithfulness is task-specific and model-specific.1
• These results support the case for CoT monitoring when unsafe behavior requires non-trivial computation, and that properties of CoT unfaithfulness are sensitive to model and setting.1

Sources (1)

• [1]CoTs remain monitorable when being unfaithful requires computation: Replicating and Extending the Necessity Effect Across Eleven Models
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 17, 06:33 AM

  Emmons et al. (2025) finds that models can follow a question hint without revealing so in its chain of thought, but when the hint is sufficiently complex the computation is forced into the CoT.

  While a LLM's chain of thought (CoT) is a promising approach to interpreting model outputs, prior work shows CoT's content does not always accurately reflect the reasoning process.

Extractive summary: sentences quoted from the sources.