Efficient LLM Safety Evaluation through Multi-Agent Debate
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_ | 1866910056855896064 |
|---|---|
| author | Lin, Dachuan Shen, Guobin Yang, Zihao Liu, Tianrong Zhao, Dongcheng Zeng, Yi |
| author_facet | Lin, Dachuan Shen, Guobin Yang, Zihao Liu, Tianrong Zhao, Dongcheng Zeng, Yi |
| contents | Safety evaluation of large language models (LLMs) increasingly relies on LLM-as-a-judge pipelines, but strong judges can still be expensive to use at scale. We study whether structured multi-agent debate can improve judge reliability while keeping backbone size and cost modest. To do so, we introduce HAJailBench, a human-annotated jailbreak benchmark with 11,100 labeled interactions spanning diverse attack methods and target models, and we pair it with a Multi-Agent Judge framework in which critic, defender, and judge agents debate under a shared safety rubric. On HAJailBench, the framework improves over matched small-model prompt baselines and prior multi-agent judges, while remaining more economical than GPT-4o under the evaluated pricing snapshot. Ablation results further show that a small number of debate rounds is sufficient to capture most of the gain. Together, these results support structured, value-aligned debate as a practical design for scalable LLM safety evaluation. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2511_06396 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Efficient LLM Safety Evaluation through Multi-Agent Debate Lin, Dachuan Shen, Guobin Yang, Zihao Liu, Tianrong Zhao, Dongcheng Zeng, Yi Artificial Intelligence Cryptography and Security I.2.7 Safety evaluation of large language models (LLMs) increasingly relies on LLM-as-a-judge pipelines, but strong judges can still be expensive to use at scale. We study whether structured multi-agent debate can improve judge reliability while keeping backbone size and cost modest. To do so, we introduce HAJailBench, a human-annotated jailbreak benchmark with 11,100 labeled interactions spanning diverse attack methods and target models, and we pair it with a Multi-Agent Judge framework in which critic, defender, and judge agents debate under a shared safety rubric. On HAJailBench, the framework improves over matched small-model prompt baselines and prior multi-agent judges, while remaining more economical than GPT-4o under the evaluated pricing snapshot. Ablation results further show that a small number of debate rounds is sufficient to capture most of the gain. Together, these results support structured, value-aligned debate as a practical design for scalable LLM safety evaluation. |
| title | Efficient LLM Safety Evaluation through Multi-Agent Debate |
| topic | Artificial Intelligence Cryptography and Security I.2.7 |
| url | https://arxiv.org/abs/2511.06396 |