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

Abstract

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

Categories

Framework mappings

@inproceedings{zeng2024how,
  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},
  booktitle = {ACL 2024},
  eprint = {2401.06373},
  archivePrefix = {arXiv},
  doi = {10.18653/v1/2024.acl-long.773},
  url = {https://arxiv.org/abs/2401.06373}
}