How Johnny Can Persuade LLMs to Jailbreak Them: Rethinking Persuasion to Challenge AI Safety

Yi Zeng, Hongpeng Lin, Jingwen Zhang, Diyi Yang, Ruoxi Jia, Weiyan Shi

Abstract

Applies social science persuasion techniques to jailbreak LLMs, showing high attack success rates using persuasion taxonomy.

Categories

Tags

Framework Mappings

Cite This Resource

@article{llmsec202400061,
  title = {How Johnny Can Persuade LLMs to Jailbreak Them: Rethinking Persuasion to Challenge AI Safety},
  author = {Yi Zeng and Hongpeng Lin and Jingwen Zhang and Diyi Yang and Ruoxi Jia and Weiyan Shi},
  year = {2024},
  journal = {ACL 2024},
  url = {https://arxiv.org/abs/2401.06373},
}