Graph of Attacks with Pruning: Optimizing Stealthy Jailbreak Prompt Generation for Enhanced LLM Content Moderation

Abstract

As large language models (LLMs) become increasingly prevalent, ensuring their robustness against adversarial misuse is crucial. This paper introduces the GAP (Graph of Attacks with Pruning) framework, an advanced approach for generating stealthy jailbreak prompts to evaluate and enhance LLM safeguards. GAP addresses limitations in existing tree-based LLM jailbreak methods by implementing an interconnected graph structure that enables knowledge sharing across attack paths. Our experimental evalua

Categories

Framework mappings

Suggested from the entry's categories.

@inproceedings{schwartz2025graph,
  title = {{Graph of Attacks with Pruning: Optimizing Stealthy Jailbreak Prompt Generation for Enhanced LLM Content Moderation}},
  author = {Daniel Schwartz and Dmitriy Bespalov and Zhe Wang and Ninad Kulkarni and Yanjun Qi},
  year = {2025},
  month = jan,
  booktitle = {Conference on Empirical Methods in Natural Language Processing},
  eprint = {2501.18638},
  archivePrefix = {arXiv},
  doi = {10.48550/arXiv.2501.18638},
  url = {https://www.semanticscholar.org/paper/79f7e65410d089da1f7a66569a19d5ff27b80f5d}
}