not aligned" is not "malicious": Being careful about hallucinations of large language models' jailbreak

Mei, Lingrui, Liu, Shenghua, Wang, Yiwei · arXiv · 2024