SRTJ: Self-Evolving Rule-Driven Training-Free LLM Jailbreaking

Abstract

LLMs are increasingly equipped with safety alignment mechanisms, yet recent studies demonstrate that they remain vulnerable to jailbreaking attacks that elicit harmful behaviors without explicit policy violations. While a growing body of work has explored automated jailbreak strategies, existing methods face several fundamental challenges, including the lack of systematic utilization of both successful and failed attack experiences, as well as the absence of principled mechanisms for composing a

Categories

Framework mappings

Suggested from the entry's categories.

@misc{li2026srtj,
  title = {{SRTJ: Self-Evolving Rule-Driven Training-Free LLM Jailbreaking}},
  author = {Jindong Li and Ying Liu and Yali Fu and Jinjing Zhu and Leyao Wang and Menglin Yang and Rex Ying},
  year = {2026},
  month = may,
  eprint = {2605.00974},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2605.00974}
}