1 paper
Yukai Zhou, Feiyang Lu, Xiaokai Mao +2
Jailbreak attacks on large language models are usually evaluated by attacker-centric metrics such as attack success rate (ASR), yet an attack that breaks a model is not necessarily…