Xiaoyi Pang, The Hong Kong University of Science and Technology; Xuanyi Hao, The State Key Laboratory of Blockchain and Data Security, Zhejiang University; Hangzhou High-Tech Zone (Binjiang) Institute of Blockchain and Data Security; Pengyu Liu, The Hong Kong University of Science and Technology; Qi Luo, Shandong University; Song Guo, The Hong Kong University of Science and Technology; Zhibo Wang, The State Key Laboratory of Blockchain and Data Security, Zhejiang University; Hangzhou High-Tech Zone (Binjiang) Institute of Blockchain and Data Security
Recent intelligent systems integrate powerful Large Language Models (LLMs) through APIs, but their trustworthiness may be critically undermined by sequence-coercing targeted attacks that force the model to emit an attacker-chosen payload, including malicious URLs, system commands, and specific misinformation. Existing defensive approaches for such threats typically rely on high access rights, impose prohibitive costs, and hinder normal inference, rendering them impractical for real-world scenarios. To solve these limitations, we introduce DualSentinel, a lightweight defense framework that can accurately and promptly detect the activation of targeted attacks alongside the LLM generation process. We first identify a characteristic of compromised LLMs, termed Entropy Lull: when a sequence-coercing targeted attack successfully hijacks the generation process, the LLM exhibits a distinct period of abnormally low and stable token probability entropy, indicating it is following a fixed path rather than making creative choices. DualSentinel leverages this pattern by developing an innovative dual-check approach. It first employs a magnitude and trend-aware monitoring method to proactively and sensitively flag an entropy lull pattern at runtime. Upon such flagging, it triggers a lightweight yet powerful secondary verification based on task-flipping. An attack is confirmed only if the entropy lull pattern persists across both the original and the flipped task, proving that the LLM's output is coercively controlled. Extensive evaluations show that DualSentinel is both highly effective (superior detection accuracy with the lowest false positives) and remarkably efficient (negligible additional cost), offering a truly practical path toward securing deployed LLMs.
USENIX Security '26 Open Access Sponsored by
King Abdullah University of Science and Technology (KAUST)
Open Access Media
USENIX is committed to Open Access to the research presented at our events. Papers and proceedings are freely available to everyone once the event begins. Any video, audio, and/or slides that are posted after the event are also free and open to everyone. Support USENIX and our commitment to Open Access.
author = {Xiaoyi Pang and Xuanyi Hao and Pengyu Liu and Qi Luo and Song Guo and Zhibo Wang},
title = {{DualSentinel}: A Lightweight Framework for Detecting Targeted Attacks in Black-box {LLM} via Dual Entropy Lull Pattern},
booktitle = {35th USENIX Security Symposium (USENIX Security 26)},
year = {2026},
isbn = {978-1-939133-58-8},
address = {Baltimore, MD},
pages = {1487--1506},
url = {https://www.usenix.org/conference/usenixsecurity26/presentation/pang-xiaoyi},
publisher = {USENIX Association},
month = aug
}