Dual-Adversarial Safety Alignment: Cultivating Intrinsic Threat Comprehension in LRMs

Abstract

Large reasoning models (LRMs) achieve remarkable success on complex tasks but remain vulnerable to harmful prompts that induce unsafe outputs. Recent methods align LRMs using direct refusals or safety rationales, yet often focus on prompt patterns rather than intrinsic attack mechanisms. As a result, these pattern-centric alignments struggle to generalize across diverse jailbreaks, compromising adversarial robustness and reasoning utility. We propose AdvSafe, a dual-adversarial framework that en

Categories

Framework mappings

Suggested from the entry's categories.

@misc{shen2026dualadversarial,
  title = {{Dual-Adversarial Safety Alignment: Cultivating Intrinsic Threat Comprehension in LRMs}},
  author = {Hongli Shen and Shaopeng Fu and Qinbo Zhang and Jian Li and Di Wang},
  year = {2026},
  month = aug,
  eprint = {2608.09542},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2608.09542}
}