作者位次 #2/7ACL 2026Findings· Findings of ACL
When Safety Alignment Fails to Generalize: Probing with Language Game Jailbreaks
Zewen Long · Yu Peng · Fangming Dong · Congyi Li · Xingmao Guan · shu wu · Kai Chen
摘要
Large language models (LLMs) are widely deployed in real-world applications, yet their safety alignment often fails to generalize beyond the specific linguistic formats seen during training. Prior work has shown that mismatched generalization can lead to alignment failures, but these studies typi...