IC-GCG: Jailbreaking Large Language Models via Intermediate Consistency Optimization

Abstract

Recent jailbreak attacks demonstrate that large language models (LLMs) can be manipulated to generate harmful outputs through adversarial prompts even after robust alignment. However, prevailing methods typically focus on forcing a desired response at the output layer—a surface-level strategy that is brittle and often fails to bypass the more fundamental safety checks embedded within the model’s internal mechanisms. In contrast, we propose intermediate consistency greedy coordinate gradient (IC-

Categories

Framework mappings

Suggested from the entry's categories.

@article{ren2026icgcg,
  title = {{IC-GCG: Jailbreaking Large Language Models via Intermediate Consistency Optimization}},
  author = {Zichu Ren and Donghai Zhu and Haibo Hong and Jun Shao},
  year = {2026},
  month = sep,
  journal = {IEEE Internet of Things Journal},
  doi = {10.1109/JIOT.2026.3707160},
  url = {https://www.semanticscholar.org/paper/7ecc8bf212ba04a3db3043938853af5bca5e6533}
}