@inproceedings{yousefpour-etal-2025-representation,
    title = "Representation Bending for Large Language Model Safety",
    author = "Yousefpour, Ashkan  and
      Kim, Taeheon  and
      Kwon, Ryan Sungmo  and
      Lee, Seungbeen  and
      Jeung, Wonje  and
      Han, Seungju  and
      Wan, Alvin  and
      Ngan, Harrison  and
      Yu, Youngjae  and
      Choi, Jonghyun",
    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.1173/",
    pages = "24073--24098",
    ISBN = "979-8-89176-251-0",
    abstract = "Large Language Models (LLMs) have emerged as powerful tools, but their inherent safety risks {--} ranging from harmful content generation to broader societal harms {--} pose significant challenges. These risks can be amplified by the recent adversarial attacks, fine-tuning vulnerabilities, and the increasing deployment of LLMs in high-stakes environments. Existing safety-enhancing techniques, such as fine-tuning with human feedback or adversarial training, are still vulnerable as they address specific threats and often fail to generalize across unseen attacks, or require manual system-level defenses. This paper introduces RepBend, a novel approach that fundamentally disrupts the representations underlying harmful behaviors in LLMs, offering a scalable solution to enhance (potentially inherent) safety. RepBend brings the idea of activation steering {--} simple vector arithmetic for steering model{'}s behavior during inference {--} to loss-based fine-tuning. Through extensive evaluation, RepBend achieves state-of-the-art performance, outperforming prior methods such as Circuit Breaker, RMU, and NPO, with up to 95{\%} reduction in attack success rates across diverse jailbreak benchmarks, all with negligible reduction in model usability and general capabilities."
}