作者位次 #2/7ACL 2026Poster· Natural Language Processing
SHARP : Self-adaptive Harmful Category-aware Prompt Generation for Black-box Jailbreaking
Yingjie Xue · Xingyou Xia · Jun Zhang · Yunbo Cao · Dengpan Ye · Guotong Geng · Fei Li
摘要
Large Language Models (LLMs) have been widely applied in various domains such as education and healthcare, making safety assurance crucial. Jailbreak attacks, a method used in red-teaming, can help evaluate and improve the defensive strategies of LLMs. However, existing jailbreak methods often ov...