作者位次 #5/6ACL 2026Poster· Natural Language Processing
Enhancing the Transferability of Jailbreak Attacks on Large Language Models via Exploiting Reparameterization Invariance
Ao Wang · Xinghao Yang · Yongshun Gong · Wei Liu · Bao-di Liu · Weifeng Liu
摘要
Jailbreak attacks serve as a pivotal technique for evaluating the safety alignment of Large language models. Current token-level attacks have shown remarkable efficacy on open-source models by leveraging gradient-based optimization. However, these attacks suffer from poor cross-model transferabil...