Diversity Helps Jailbreak Large Language Models
Weiliang Zhao, Daniel Ben-Levi, Wei Hao, Junfeng Yang, Chengzhi Mao
Abstract
We have uncovered a powerful jailbreak technique that leverages large language models’ ability to diverge from prior context, enabling them to bypass safety constraints and generate harmful outputs. By simply instructing the LLM to deviate and obfuscate previous attacks, our method dramatically outperforms existing approaches, achieving up to a 62.83% higher success rate in compromising ten leading chatbots, including GPT-4, Gemini, and Llama, while using only 12.9% of the queries. This revelation exposes a critical flaw in current LLM safety training, suggesting that existing methods may merely mask vulnerabilities rather than eliminate them. Our findings sound an urgent alarm for the need to revolutionize testing methodologies to ensure robust and reliable LLM security.
BibTeX
@inproceedings{zhao-etal-2025-diversity,
title = "Diversity Helps Jailbreak Large Language Models",
author = "Zhao, Weiliang and
Ben-Levi, Daniel and
Hao, Wei and
Yang, Junfeng and
Mao, Chengzhi",
editor = "Chiruzzo, Luis and
Ritter, Alan and
Wang, Lu",
booktitle = "Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)",
month = apr,
year = "2025",
address = "Albuquerque, New Mexico",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.naacl-long.238/",
pages = "4647--4680",
ISBN = "979-8-89176-189-6"
}