作者位次 #7/10ACL 2026Poster· Natural Language Processing
Into the Gray Zone: Domain Contexts Can Blur LLM Safety Boundaries
Ki Sen Hung · Xi Yang · Chang Liu · Haoran Li · Kejiang Chen · Changxuan Fan · Tsun On Kwok · Weiming Zhang · Xiaomeng Li · Yangqiu Song
摘要
A central goal of LLM alignment is to balance helpfulness with harmlessness, yet these objectives conflict when the same knowledge serves both legitimate and malicious purposes. This tension is amplified by context-sensitive alignment: we observe that domain-specific contexts (e.g., chemistry) se...