# Are Reasoning LLMs Robust to Interventions on Their Chain-of-Thought?

arXiv: https://arxiv.org/abs/2602.07470 · HTML: https://mlsanity.com/paper/2602.07470/

## Key facts

- Cited by 6 papers in the map (data through 14 Oct 2026).
- Placed in Mathematical reasoning, one of 17 directions in LLM reasoning.
- Builds on 1 of its 55 references that are in the map.
- First posted on arXiv 7 Feb 2026.

## Placed in

- LLM reasoning / [Mathematical reasoning](https://mlsanity.com/lr/mathematical-reasoning/)

## Builds on (most cited first, 1 of 1)

- [Prompt Injection attack against LLM-integrated Applications](https://mlsanity.com/paper/2306.05499/) (2023)

## Built on by (most cited first, 6 of 6)

- [Evaluation Awareness in Language Models Has Limited Effect on Behaviour](https://mlsanity.com/paper/2605.05835/) (2026)
- [Robust Reasoning Benchmark](https://mlsanity.com/paper/2604.08571/) (2026)
- [Reliable Chain-of-Thought via Prefix Consistency](https://mlsanity.com/paper/2605.07654/) (2026)
- [SABER: Stability-Aware Early Exit for LLM Reasoning via Adversarial Branch Probing](https://mlsanity.com/paper/2608.27963/) (2026)
- [Should We be Pedantic About Reasoning Errors in Machine Translation?](https://mlsanity.com/paper/2604.09890/) (2026)
- [Adaptive Triggering for Bias Correction in LLM Reasoning](https://mlsanity.com/paper/2608.25379/) (2026)

## Abstract (arXiv)

Reasoning LLMs (RLLMs) generate step-by-step chains of thought (CoTs) before giving an answer, which improves performance on complex tasks and makes reasoning more transparent. But how robust are these reasoning traces to disruptions that occur within them? To address this question, we introduce a controlled evaluation framework that perturbs a model's own CoT at fixed timesteps. We design seven interventions (benign, neutral, and adversarial) and apply them to multiple open-weight RLLMs across Math, Science, and Logic tasks. Our results show that RLLMs are generally robust, reliably recovering from diverse perturbations, with robustness improving with model size and degrading when interventions occur early. However, robustness is not style-invariant: paraphrasing suppresses doubt-like expressions and reduces performance, while other interventions trigger doubt and support recovery. Recovery also carries a cost: neutral and adversarial noise can inflate CoT length by more than 200%, whereas paraphrasing shortens traces but harms accuracy. These findings provide new evidence on how RLLMs maintain reasoning integrity, identify doubt as a central recovery mechanism, and highlight trade-offs between robustness and efficiency that future training methods should address.