arXiv · ML· Zixuan Wang, Hao Li, Fengyu Gao, G. Edward Suh, Yi Zeng, Yevgeniy Vorobeychik, Ning Zhang, Chaowei Xiao·· 2026-10-06AI 评分35
Self-Reflection Fine-Tuning (SRFT) 增强 LLM 智能体抵御提示注入攻击的鲁棒性
Self-Reflection Fine-Tuning: Enhancing Agent Security against Prompt Injection Attacks from Failure Experience
AI 导读
提出 Self-Reflection Fine-Tuning(SRFT)训练框架,利用 LLM 智能体在对抗条件下的失败经验生成结构化自反思推理,显著提升模型对提示注入攻击的鲁棒性。
来源:arXiv · ML · arxiv.org