Learning-Based Automated Adversarial Red-Teaming for Robustness Evaluation of Large Language Models
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866915961709264896 |
|---|---|
| author | Wei, Zhang Chen, Hanxuan Hu, Peilu Wei, Zhenyuan Liang, Chenwei Luo, Jing Ni, Ziyi Yan, Hao Mei, Li Lang, Shengning Lu, Kuan Xiao, Xi Han, Zhimo Wang, Yijin Zhang, Yichao Yang, Chen Hao, Junfeng Gu, Jiayi Bao, Riyang Wang, Mu-Jiang-Shan |
| author_facet | Wei, Zhang Chen, Hanxuan Hu, Peilu Wei, Zhenyuan Liang, Chenwei Luo, Jing Ni, Ziyi Yan, Hao Mei, Li Lang, Shengning Lu, Kuan Xiao, Xi Han, Zhimo Wang, Yijin Zhang, Yichao Yang, Chen Hao, Junfeng Gu, Jiayi Bao, Riyang Wang, Mu-Jiang-Shan |
| contents | The increasing deployment of large language models (LLMs) in safety-critical applications raises fundamental challenges in systematically evaluating robustness against adversarial behaviors. Existing red-teaming practices are largely manual and expert-driven, which limits scalability, reproducibility, and coverage in high-dimensional prompt spaces. We formulate automated LLM red-teaming as a structured adversarial search problem and propose a learning-driven framework for scalable vulnerability discovery. The approach combines meta-prompt-guided adversarial prompt generation with a hierarchical execution and detection pipeline, enabling standardized evaluation across six representative threat categories, including reward hacking, deceptive alignment, data exfiltration, sandbagging, inappropriate tool use, and chain-of-thought manipulation. Extensive experiments on GPT-OSS-20B identify 47 vulnerabilities, including 21 high-severity failures and 12 previously undocumented attack patterns. Compared with manual red-teaming under matched query budgets, our method achieves a 3.9$\times$ higher discovery rate with 89\% detection accuracy, demonstrating superior coverage, efficiency, and reproducibility for large-scale robustness evaluation. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2512_20677 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Learning-Based Automated Adversarial Red-Teaming for Robustness Evaluation of Large Language Models Wei, Zhang Chen, Hanxuan Hu, Peilu Wei, Zhenyuan Liang, Chenwei Luo, Jing Ni, Ziyi Yan, Hao Mei, Li Lang, Shengning Lu, Kuan Xiao, Xi Han, Zhimo Wang, Yijin Zhang, Yichao Yang, Chen Hao, Junfeng Gu, Jiayi Bao, Riyang Wang, Mu-Jiang-Shan Cryptography and Security Computation and Language The increasing deployment of large language models (LLMs) in safety-critical applications raises fundamental challenges in systematically evaluating robustness against adversarial behaviors. Existing red-teaming practices are largely manual and expert-driven, which limits scalability, reproducibility, and coverage in high-dimensional prompt spaces. We formulate automated LLM red-teaming as a structured adversarial search problem and propose a learning-driven framework for scalable vulnerability discovery. The approach combines meta-prompt-guided adversarial prompt generation with a hierarchical execution and detection pipeline, enabling standardized evaluation across six representative threat categories, including reward hacking, deceptive alignment, data exfiltration, sandbagging, inappropriate tool use, and chain-of-thought manipulation. Extensive experiments on GPT-OSS-20B identify 47 vulnerabilities, including 21 high-severity failures and 12 previously undocumented attack patterns. Compared with manual red-teaming under matched query budgets, our method achieves a 3.9$\times$ higher discovery rate with 89\% detection accuracy, demonstrating superior coverage, efficiency, and reproducibility for large-scale robustness evaluation. |
| title | Learning-Based Automated Adversarial Red-Teaming for Robustness Evaluation of Large Language Models |
| topic | Cryptography and Security Computation and Language |
| url | https://arxiv.org/abs/2512.20677 |