How Jailbreak Attacks Inform Safety Alignment: A Defender-Centric, Shapley-Based Evaluation of Jailbreak Contributions

Abstract

Jailbreak attacks on large language models are usually evaluated by attacker-centric metrics such as attack success rate (ASR), yet an attack that breaks a model is not necessarily useful for improving its safety. We propose a defender-centric view of jailbreak evaluation, where attacks are evaluated by the downstream safety improvements they enable when used as red-teaming data for safety training. Building on this view, we introduce A-MESS (Minimal Effective Attack-Subset Selection), a setting

Categories

Framework mappings

Suggested from the entry's categories.

@misc{zhou2026how,
  title = {{How Jailbreak Attacks Inform Safety Alignment: A Defender-Centric, Shapley-Based Evaluation of Jailbreak Contributions}},
  author = {Yukai Zhou and Feiyang Lu and Xiaokai Mao and Jinfei Liu and Wenjie Wang},
  year = {2026},
  month = jul,
  eprint = {2607.17152},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2607.17152}
}