ReasoningGuard: Safeguarding Large Reasoning Models with Inference-time Safety Aha Moments

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wang, Yuquan, Zhang, Mi, Wang, Yining, Hong, Geng, Wen, Mi, You, Xiaoyu, Yang, Min
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917462661922816
author Wang, Yuquan
Zhang, Mi
Wang, Yining
Hong, Geng
Wen, Mi
You, Xiaoyu
Yang, Min
author_facet Wang, Yuquan
Zhang, Mi
Wang, Yining
Hong, Geng
Wen, Mi
You, Xiaoyu
Yang, Min
contents Large Reasoning Models (LRMs) have demonstrated impressive performance in reasoning-intensive tasks, but they remain vulnerable to harmful content generation, particularly in the mid-to-late steps of their reasoning processes. Current defense methods, however, depend on costly fine-tuning and additional expert knowledge, which limits their scalability. In this work, we propose ReasoningGuard, an inference-time safeguard for LRMs. It injects timely safety aha moments during the reasoning process to guide the model towards harmless yet helpful reasoning. Our approach leverages the internal attention mechanisms of the LRM to accurately identify key points in the reasoning path, triggering safety-oriented reflections. To safeguard both the subsequent reasoning steps and the final answers, we implement a scaling sampling strategy during decoding to select the optimal reasoning path. With minimal additional inference cost, ReasoningGuard effectively mitigates four types of jailbreak attacks, including recent ones targeting the reasoning process of LRMs. Our approach outperforms nine existing safeguards, providing state-of-the-art defenses while avoiding common exaggerated safety issues.
format Preprint
id arxiv_https___arxiv_org_abs_2508_04204
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle ReasoningGuard: Safeguarding Large Reasoning Models with Inference-time Safety Aha Moments
Wang, Yuquan
Zhang, Mi
Wang, Yining
Hong, Geng
Wen, Mi
You, Xiaoyu
Yang, Min
Computation and Language
Artificial Intelligence
Large Reasoning Models (LRMs) have demonstrated impressive performance in reasoning-intensive tasks, but they remain vulnerable to harmful content generation, particularly in the mid-to-late steps of their reasoning processes. Current defense methods, however, depend on costly fine-tuning and additional expert knowledge, which limits their scalability. In this work, we propose ReasoningGuard, an inference-time safeguard for LRMs. It injects timely safety aha moments during the reasoning process to guide the model towards harmless yet helpful reasoning. Our approach leverages the internal attention mechanisms of the LRM to accurately identify key points in the reasoning path, triggering safety-oriented reflections. To safeguard both the subsequent reasoning steps and the final answers, we implement a scaling sampling strategy during decoding to select the optimal reasoning path. With minimal additional inference cost, ReasoningGuard effectively mitigates four types of jailbreak attacks, including recent ones targeting the reasoning process of LRMs. Our approach outperforms nine existing safeguards, providing state-of-the-art defenses while avoiding common exaggerated safety issues.
title ReasoningGuard: Safeguarding Large Reasoning Models with Inference-time Safety Aha Moments
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2508.04204