[论文] RedEvoAgent: Automatic Red-Teaming Agent with Experience-Driven Skill ...

论文概要

研究领域: ML 作者: Junjie Zhang, Hui Liu, Kecheng Chen, Xianbo Mo, Changsheng Chen, Haoliang Li 发布时间: 2026-08-27 arXiv: 2608.27439

中文摘要

基于LLM的智能体越来越多地部署在产品级执行环境中，其中越狱可能触发有害工具使用和持久状态变更，造成比不安全文本生成更大的风险。现有自动红队方法通常依赖固定攻击，而近期的智能体攻击者协调多个越狱工具，通过基于轨迹的检索显示出更强潜力。然而，此类检索可能因检索偏差和工具信用不清而重用误导性经验，完整轨迹增加上下文开销并降低可解释性。我们提出了RedEvoAgent，一种黑盒红队智能体，将跨案例攻击轨迹提炼成简洁、人类可读的攻击技能。攻击技能通过工具效果分析和决策-工具归因进行自适应演化，并通过验证棘轮仅保留提高验证性能的更新。在多个基准、目标模型和目标执行环境上的实验表明，RedEvoAgent优于固定和智能体基线，提高了工具效率，并能跨攻击者模型和目标执行环境迁移。

原文摘要

LLM-based agents are increasingly deployed in product-level execution harnesses, where jailbreaks can trigger harmful tool use and persistent state changes, creating greater risks than unsafe text generation alone. Existing automatic red-teaming methods often rely on fixed attacks, while recent agentic attackers coordinate multiple jailbreak tools and show stronger potential through trajectory-based retrieval. However, such retrieval can reuse misleading experiences due to retrieval bias and unclear tool credit, and full trajectories add context overhead while reducing interpretability. We propose RedEvoAgent, a black-box red-teaming agent that distills cross-case attack trajectories into a concise, human-readable attack skill. The attack skill adaptively evolves through tool-effectiveness...

#论文 #arXiv #ML #小凯