@inproceedings{yuan-etal-2025-refuse,
    title = "Refuse Whenever You Feel Unsafe: Improving Safety in {LLM}s via Decoupled Refusal Training",
    author = "Yuan, Youliang  and
      Jiao, Wenxiang  and
      Wang, Wenxuan  and
      Huang, Jen-tse  and
      Xu, Jiahao  and
      Liang, Tian  and
      He, Pinjia  and
      Tu, Zhaopeng",
    editor = "Che, Wanxiang  and
      Nabende, Joyce  and
      Shutova, Ekaterina  and
      Pilehvar, Mohammad Taher",
    booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = jul,
    year = "2025",
    address = "Vienna, Austria",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingestion-acl-25/2025.acl-long.158/",
    pages = "3149--3167",
    ISBN = "979-8-89176-251-0",
    abstract = "This study addresses a critical gap in safety tuning practices for Large Language Models (LLMs) by identifying and tackling a refusal position bias within safety tuning data, which compromises the models' ability to appropriately refuse generating unsafe content. We introduce a novel approach, \textbf{De}coupled \textbf{R}efusal \textbf{T}r\textbf{a}ining (DeRTa), designed to empower LLMs to refuse compliance to harmful prompts at any response position, significantly enhancing their safety capabilities. DeRTa incorporates two novel components: (1) Maximum Likelihood Estimation (MLE) with Harmful Response Prefix, which trains models to recognize and avoid unsafe content by appending a segment of harmful response to the beginning of a safe response, and (2) Reinforced Transition Optimization (RTO), which equips models with the ability to transition from potential harm to safety refusal consistently throughout the harmful response sequence. Our empirical evaluation, conducted using LLaMA3 and Mistral model families across six attack scenarios, demonstrates that our method not only improves model safety without compromising performance but also surpasses baseline methods in defending against attacks."
}