AdversaBench: 基于多裁判确认与跨模型迁移性的自动化大语言模型红队测试
AdversaBench: Automated LLM Red-Teaming with Multi-Judge Confirmation and Cross-Model Transferability
AI总结：
提出AdversaBench端到端红队测试流水线，使用五种结构化算子变异种子提示，通过三裁判加元裁判机制确认失败，实验发现算子效果因类别而异、二元失败率掩盖难度、裁判一致性受标签偏斜影响、对抗提示可跨模型迁移。
摘要：
扩展大语言模型的对抗性评估需要一种生成困难输入的方法以及一种可靠的方式来确认由此产生的失败是真实的。我们提出了AdversaBench，一个端到端的红队测试流水线，它使用五种结构化算子变异种子提示，查询目标模型，并通过一个三裁判小组（附带元裁判打破平局）确认失败。我们报告了在三个类别（推理、指令遵循和工具使用）中45个种子上的实验。每个种子都产生了确认的失败。四个发现尤为突出。第一，算子有效性因类别而异：inject_distractor在指令遵循种子上的平均奖励为0.00，但在推理和工具使用上为0.80-0.83。第二，二元失败率掩盖了难度：指令遵循种子平均需要2.4次攻击者迭代，而其他类别为1.1次，这一差距在生存曲线中可见。第三，成对裁判一致性为80-87%，但由于标签偏斜，Cohen's kappa接近零；类别级别的不一致率更具信息量。第四，针对Llama 3.1 8B生成的对抗提示可以零样本迁移到Llama 3.3 70B，这表明这些变异利用了通用行为模式而非模型特定的弱点。代码、数据集和分析脚本可在https://this URL获取。
英文摘要：
Scaling adversarial evaluation of large language models requires both a method for generating hard inputs and a reliable way to confirm that resulting failures are real. We present AdversaBench, an end-to-end red-teaming pipeline that mutates seed prompts with five structured operators, queries a target model, and confirms failures through a three-judge panel with a meta-judge tiebreaker. We report experiments on 45 seeds across three categories: reasoning, instruction-following, and tool use. Every seed produced a confirmed failure. Four findings stand out. First, operator effectiveness varies sharply by category: inject_distractor scores 0.00 mean reward on instruction-following seeds but 0.80-0.83 on reasoning and tool-use. Second, binary failure rate hides difficulty: instruction-following seeds required 2.4 attacker iterations on average versus 1.1 for other categories, a gap visible in survival curves. Third, pairwise judge agreement of 80-87% coexists with near-zero Cohen's kappa due to label skew; category-level disagreement rates are more informative. Fourth, adversarial prompts generated against Llama 3.1 8B transfer zero-shot to Llama 3.3 70B, suggesting the mutations exploit general behavioral patterns rather than model-specific weaknesses. Code, dataset, and analysis scripts are available at https://github.com/khanak0509/AdversaBench .
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
