← Publications
ICML 2025 · New In MLWorkshop poster

BlackDAN: A Black-Box Multi-Objective Approach for Effective and Contextual Jailbreaking of Large Language Models

Xinyuan Wang, Victor Shea-Jay Huang, Renmiao Chen, Hao Wang, Chengwei Pan, Lei Sha, Minlie Huang

ICML arXiv Code Citation
BlackDAN's framework combines black-box model responses, judge models, and NSGA-II multi-objective optimization.
BlackDAN's framework combines black-box model responses, judge models, and NSGA-II multi-objective optimization.

TL;DR

A black-box evaluation framework uses multi-objective optimization to study jailbreak success, contextual relevance, and detectability together.以多目标优化联合考察越狱成功率、上下文相关性与可检测性，构建黑盒大语言模型安全评测框架。

Citation

Download .bib
@misc{wang2024blackdan,
  title = {{BlackDAN}: A Black-Box Multi-Objective Approach for Effective and Contextual Jailbreaking of Large Language Models},
  author = {Wang, Xinyuan and Huang, Victor Shea-Jay and Chen, Renmiao and Wang, Hao and Pan, Chengwei and Sha, Lei and Huang, Minlie},
  year = {2024},
  eprint = {2410.09804},
  archivePrefix = {arXiv},
  primaryClass = {cs.CR},
  note = {Presented at the ICML 2025 New In ML affinity workshop},
  url = {https://arxiv.org/abs/2410.09804}
}