INJECAGENT
收藏资源简介:
INJECAGENT数据集由伊利诺伊大学厄巴纳-香槟分校的研究团队开发,旨在评估和防御集成工具的大型语言模型代理面临的间接提示注入攻击。该数据集包含1054个测试案例,涵盖17种不同的用户工具和62种攻击者工具,用于模拟和分析攻击场景。数据集通过GitHub平台共享,支持研究者评估和提升语言模型代理的安全性,特别是在处理外部内容时的风险管理。
INJECAGENT Dataset is developed by the research team from the University of Illinois Urbana-Champaign, with the goal of evaluating and defending against indirect prompt injection attacks targeting large language model (LLM) agents integrated with external tools. This dataset comprises 1054 test cases, covering 17 distinct user tools and 62 attacker tools, to simulate and analyze various attack scenarios. The dataset is shared through the GitHub platform, enabling researchers to assess and improve the security of LLM agents, especially for risk management when processing external content.




