@inproceedings{tong-etal-2026-rlshield,
    title = "{RLS}hield: Dynamic Jailbreak Detection for {LLM}s via Reinforced Adaptive Learning",
    author = "Tong, Zhao  and
      Yang, Pengfei  and
      Gu, Yimeng  and
      Shi, Haichao  and
      Liu, Qiang  and
      Xu, Xingcheng  and
      Wu, Shu  and
      Zhang, Xiao-Yu",
    editor = "Liakata, Maria  and
      Moreira, Viviane P.  and
      Zhang, Jiajun  and
      Jurgens, David",
    booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
    month = jul,
    year = "2026",
    address = "San Diego, California, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl/2026.findings-acl.1182/",
    pages = "23615--23630",
    ISBN = "979-8-89176-395-1",
    abstract = "While prompt engineering enhances the capabilities of Large Language Models (LLMs), it also exposes critical safety concerns. Due to the inherent brittleness of their static safety boundaries, LLMs are vulnerable to jailbreak prompts, i.e. adversarial inputs designed to bypass safeguards and induce the generation of harmful content. Existing detection mechanisms rely on static model components or fixed decision thresholds, limiting their ability to generalize to evolving attack patterns and continual model updates. To bridge this gap, we propose RLShield, a dynamic jailbreak detection framework that employs reinforcement learning for adaptive threshold selection. RLShield incorporates three key innovations: (i) a dynamic retrieval and LLM-based rewriting module to simulate diverse adversarial contexts; (ii) a cross-layer representation analysis to pinpoint safety-critical parameters; and (iii) a Soft Actor-Critic (SAC) based agent that learns to predict optimal, sample-specific detection thresholds. Experimental results demonstrate that RLShield consistently outperforms state-of-the-art baselines in detection performance while maintaining high computational efficiency. Notably, it improves F1 by up to 7.3{\%}, while achieving an average of 3$\times$ gain in inference efficiency across multiple LLM backbones."
}