Hugging Face Papers ↗ · 2026-11-18
PIPETTE:针对提示注入的紫队测试以提升大型语言模型代理的鲁棒性
媒体报道 Hugging Face Papers 原文语言:中文
研究提出了 PIPETTE 框架,通过字典学习和在策略对抗训练,克服了现有基于强化学习的红队方法的局限,提升了大型语言模型代理对间接提示注入攻击的鲁棒性。
全文
A novel framework addresses the limitations of current reinforcement learning-based red-teaming methods by employing dictionary learning and on-policy adversarial training to improve the robustness of LLM agents against indirect prompt injection attacks.
正文以来源语言显示。