作者位次 #2/10ACL 2026Poster· Natural Language Processing
Probing the Safety Robustness of LLM s in Latent Space
Tianle Gu · Kexin Huang · Zongqi Wang · Yixu Wang · Jie Li · Xin Wang · Yang Yao · Yujiu Yang · Yan Teng · Yingchun Wang
摘要
Safety alignment is a fundamental prerequisite for building trustworthy artificial general intelligence. Despite substantial progress in safety alignment techniques, empirical evidence shows that aligned large language models can still produce unsafe responses under minor internal perturbations,...