ACL Findings2026
RLShield: Dynamic Jailbreak Detection for LLMs via Reinforced Adaptive Learning
Zhao Tong, Pengfei Yang, Yimeng Gu, Haichao Shi, Qiang Liu, Xingcheng Xu, Shu Wu, Xiao-Yu Zhang
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 crosslayer representation analysis to pinpoint safetycritical 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× gain in inference efficiency across multiple LLM backbones. Our codes are available at this website.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
Lineage
Builds on19
- Gradient Cuff: Detecting Jailbreak Attacks on Large Language Models by Exploring Refusal Loss LandscapesXiaomeng Hu, Pin-Yu Chen, Tsung-Yi HoNeurIPS 2024 · 97 citations
- On the Risk of Misinformation Pollution with Large Language ModelsYikang Pan, Liangming Pan, Wenhu Chen, Preslav Nakov et al.EMNLP Findings 2023 · 85 citations
- ToxicChat: Unveiling Hidden Challenges of Toxicity Detection in Real-World User-AI ConversationZi Lin, Zihan Wang, Yongqi Tong, Yangkun Wang et al.EMNLP Findings 2023 · 44 citations
- Defending LLMs against Jailbreaking Attacks via BacktranslationYihan Wang, Zhouxing Shi, Andrew Bai, Cho-Jui HsiehACL Findings 2024 · 28 citations
- Generate First, Then Sample: Enhancing Fake News Detection with LLM-Augmented Reinforced SamplingZhao Tong, Yimeng Gu, Huidong Liu, Qiang Liu et al.ACL 2025 · 14 citations
Related papers
- JBShield: Defending Large Language Models from Jailbreak Attacks through Activated Concept Analysis and ManipulationShenyi Zhang, Yuchen Zhai, Keyan Guo, Hongxin Hu et al.USENIX Security 2025
- GraphShield: Graph-Theoretic Modeling of Network-Level Dynamics for Robust Jailbreak DetectionSunghee Dong, Sungwon Yi, Kangmin Bae, Jaeyoon Kim et al.ICLR 2026
- MirrorShield: Towards Dynamic Adaptive Defense Against Jailbreaks via Entropy-Guided Mirror CraftingRui Pu, Chaozhuo Li, Rui Ha, Litian Zhang et al.AAAI 2026
- Auto-RT: Automatic Jailbreak Strategy Exploration for Red-Teaming Large Language ModelsYanjiang Liu, Shuheng Zhou, Yaojie Lu, Huijia Zhu et al.ICLR 2026 · 10 citations
- Sugar-Coated Poison: Benign Generation Unlocks JailbreakingYuhang Wu, Yu-Jie Xiong, Hao Zhang, Jia-Chen Zhang et al.EMNLP Findings 2025 · 3 citations
