Reasoning Models Don't Always Say What They Think

By Yanda Chen, Joe Benton, Ansh Radhakrishnan, Jonathan Uesato, Carson E. Denison, John Schulman, Arushi Somani, P. Hase, Misha Wagner, Fabien Roger and 5 more. arXiv.org, 2025

366 citations recorded by the source corpus.

Chain-of-thought (CoT) offers a potential boon for AI safety as it allows monitoring a model's CoT to try to understand its intentions and reasoning processes. However, the effectiveness of such monitoring hinges on CoTs faithfully representing models' actual reasoning processes. We evaluate CoT faithfulness of state-of-the-art reasoning models across 6 reasoning hints presented in the prompts and find: (1) for most settings and models tested, CoTs reveal their usage of hints in at least 1% of examples where they use the hint, but the reveal rate is often below 20%, (2) outcome-based reinforcement learning initially improves faithfulness but plateaus without saturating, and (3) when reinforcement learning increases how frequently hints are used (reward hacking), the propensity to verbalize them does not increase, even without training against a CoT monitor. These results suggest that CoT monitoring is a promising way of noticing undesired behaviors during training and evaluations, but that it is not sufficient to rule them out. They also suggest that in settings like ours where CoT reasoning is not necessary, test-time monitoring of CoTs is unlikely to reliably catch rare and catastrophic unexpected behaviors.

Subfields: Artificial Intelligence, Safety Research

View Paper: https://www.semanticscholar.org/paper/b9ca6db27f02a9ddf0d4fdb51b26432c99a27be0

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

  1. Mehrdad Farajtabar
  2. Keivan Alizadeh-Vahid
  3. Samy Bengio
  4. Yoshua Bengio
  5. Sergey Levine
  6. Jun Zhu
  7. Quoc V. Le
  8. Mohit Bansal

Citations in the corpus