Attention-Guided Reward for Reinforcement Learning-based Jailbreak against Large Reasoning Models
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866914579570753536 |
|---|---|
| author | Lin, Zheng Niu, Zhenxing Ji, Haoxuan Huang, Yuzhe Gao, Haichang |
| author_facet | Lin, Zheng Niu, Zhenxing Ji, Haoxuan Huang, Yuzhe Gao, Haichang |
| contents | Large Reasoning Models (LRMs) have demonstrated remarkable capabilities in solving complex problems by generating structured, step-by-step reasoning content. However, exposing a model's internal reasoning process introduces additional safety risks; for example, recent studies show that LRMs are more vulnerable to jailbreak attacks than standard LLMs. In this paper, we investigate jailbreak attacks on LRMs and reveal that the attack success rate (ASR) is closely correlated with LRMs' attention patterns. Specifically, successful jailbreaks tend to assign lower attention to harmful tokens in the input prompt, while allocating higher attention to those tokens in the reasoning content. Motivated by this finding, we propose a novel jailbreak method for LRMs that leverages reinforcement learning (RL) to enhance attack effectiveness, explicitly incorporating attention signals into the reward function design. In addition, we introduce diverse persuasion strategies to enrich the RL action space, which consistently improves the ASR. Extensive experiments on five open-source and closed-source LRMs across three benchmarks demonstrate that our method achieves substantially higher ASR, outperforming existing approaches in terms of effectiveness, efficiency, and transferability. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_19485 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Attention-Guided Reward for Reinforcement Learning-based Jailbreak against Large Reasoning Models Lin, Zheng Niu, Zhenxing Ji, Haoxuan Huang, Yuzhe Gao, Haichang Artificial Intelligence Large Reasoning Models (LRMs) have demonstrated remarkable capabilities in solving complex problems by generating structured, step-by-step reasoning content. However, exposing a model's internal reasoning process introduces additional safety risks; for example, recent studies show that LRMs are more vulnerable to jailbreak attacks than standard LLMs. In this paper, we investigate jailbreak attacks on LRMs and reveal that the attack success rate (ASR) is closely correlated with LRMs' attention patterns. Specifically, successful jailbreaks tend to assign lower attention to harmful tokens in the input prompt, while allocating higher attention to those tokens in the reasoning content. Motivated by this finding, we propose a novel jailbreak method for LRMs that leverages reinforcement learning (RL) to enhance attack effectiveness, explicitly incorporating attention signals into the reward function design. In addition, we introduce diverse persuasion strategies to enrich the RL action space, which consistently improves the ASR. Extensive experiments on five open-source and closed-source LRMs across three benchmarks demonstrate that our method achieves substantially higher ASR, outperforming existing approaches in terms of effectiveness, efficiency, and transferability. |
| title | Attention-Guided Reward for Reinforcement Learning-based Jailbreak against Large Reasoning Models |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2605.19485 |