RouteHijack: Routing-Aware Attack on Mixture-of-Experts LLMs

Abstract

Safety alignment is critical for the responsible deployment of large language models (LLMs). As Mixture-of-Experts (MoE) architectures are increasingly adopted to scale model capacity, understanding their safety robustness becomes essential. Existing adversarial attacks, however, have notable limitations. Prompt-based jailbreaks rely on heuristic search and transfer poorly, model intervention methods require privileged access to internal representations, and optimization-based input attacks rema

Categories

Framework mappings

Suggested from the entry's categories.

@misc{xu2026routehijack,
  title = {{RouteHijack: Routing-Aware Attack on Mixture-of-Experts LLMs}},
  author = {Zhiyuan Xu and Joseph Gardiner and Sana Belguith and Lichao Wu},
  year = {2026},
  month = may,
  eprint = {2605.02946},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2605.02946}
}