📢 USENIX Security 2026 我实验室将有 8 篇论文在会上进行分享！ 🎉🎉🎉🎉🎉🎉🎉🎉🎉🎉
📌 USENIX Security Symposium 是系统安全领域四大顶会之一，与 S&P、CCS、NDSS 并列它不仅是 CCF-A 类会议，更是每年安全研究者们最关注的学术舞台之一。
👀 想知道我们都中了哪些论文？马上来一波论文大公开👇👇👇
Towards Generality: Task-Adaptive Binary Analysis via Semantic Retrieval and Verifiable Reasoning
Authors:
Yuzhe Liu, Zhijie Liu, Zhengmin Yu, Shu Wang, Ling Jiang, Sen Nie, Shi Wu, Zhanyong Tang, and Yuan Zhang
Abstract:
Stripped binaries dominate real-world security analysis: COTS software, firmware, and malware are released without symbols, obscuring program semantics. Recent LLM-based approaches target narrowly predefined tasks, limiting applicability in intent-driven analysis. Toward general binary analysis, agentic LLMs provide a natural direction, yet achieving generality remains challenging: query intent is hard to ground in stripped binaries, and effective tool-driven analysis is difficult.
We present BINREX, the first agentic framework for general, fully automated static binary analysis. BINREX employs a dual-encoder pipeline to enrich stripped functions with semantic context, paired with hierarchical planning to decompose intents into executable subtasks. It leverages code synthesis to translate subtasks into IDAPython scripts, executed with iterative validation to produce outputs aggregated into the final report.
We evaluate BINREX on BinREval, a benchmark of 72 tasks across seven COTS categories with machine-checkable oracles. BINREX achieves 83.3% success rate, outperforming baselines (Codex: 27.8%, Codex with IDA: 43.1%) while reducing total analysis time from 87.9 to 36.1 hours. On domain benchmarks (Juliet, NYU CTF), BINREX is competitive with task-specific methods on their respective evaluation settings. Human study and industrial deployment confirm practical value: BINREX identifies 395 unknown malware samples with 96× average efficiency gain over experts.
KernelRCA: Facilitating Root Cause Analysis of Memory Corruptions in Linux Kernel with Contextual Causality Chain
Authors:
Kangzheng Gu, Yifan Zhang, Yuan Zhang, and Min Yang
Abstract:
Continuous fuzzing infrastructure has found a large number of bugs. In this case, automatic root cause analysis (RCA) has been proposed to reduce the expensive manual effort to understand the root cause of a bug. However, existing root-cause representations are designed as isolated forms. Analysts still need to manually infer the integrated bug-triggering procedure including calling context and data dependency, which is very difficult for OS kernels due to their complexity. In this paper, we propose contextual causality chain (CCchain), a novel root-cause representation to intuitively reflect the integrated bug-triggering procedure of memory corruptions in the Linux kernel. CC-chain shows the bugcontributing instructions to explain corresponding unexpected behaviors that lead to a bug, as well as calling contexts and data dependencies among these instructions to help analysts rapidly understand how a bug happens. To automatically construct the CC-chain, we design a root cause analysis system KernelRCA including selective tracing, contextual information recovery, and chain-style root cause analysis. KernelRCA successfully diagnoses 54 various kinds of real-world memory corruptions in the Linux kernel and performs better than existing crash reports and KASAN reports. A user study shows that KernelRCA’s reports significantly facilitate bug understanding and fixing for human analysts.
Khost: KVM-based Near Native MCU Firmware Rehosting
Authors:
Chunlin Wang, Yicheng Yang, Yuan Zhang, Haoyu Xiao, Yifan Zhang, and Jiarun Dai
Abstract:
Microcontroller Unit (MCU)-based devices constitute a critical layer of the Internet of Things (IoT) infrastructure, so ensuring their security is of paramount importance. Rehostingbased dynamic MCU firmware analysis is an effective approach to securing these devices. However, existing rehosting frameworks commonly suffer from substantial performance overhead due to emulation or diminished execution scope. To address these limitations, we propose Khost, a nearnative, scope-preserving rehosting framework. It extends the KVM by introducing a lightweight extended CPU, an auxiliary page table, and a software-based interrupt controller, enabling MCU firmware to be rehosted on high-performance platforms with minimum overhead. It also provides a memorymapped I/O (MMIO) monitor for quick peripheral interactions and a wrapper for firmware to enable coverage collection and configure the existing fuzzing engines flexibly. Evaluations on two standard benchmarks show that Khost reduces overhead by 90.0% to 95.5% for complex computational tasks and by up to 98.5% for MCU system-level operations, compared to QEMU. Furthermore, fuzzing on 12 real-world firmware with Khost achieves up to 197.5× higher throughput and improves basic block coverage by 6x compared to existing fuzzing tools. Additionally, Khost successfully uncovers 5 previously unknown bugs.
Patch-Guided Vulnerability Detection: Extracting Java API Security Rules via Attack–Defense Cross-Analysis
Authors:
Bofei Chen, Shuang Liao, Lei Zhang, Chibin Zhang, Mathias Payer, and Yuan Zhang
Abstract:
Security-sensitive APIs are critical components in modern Java applications, yet improper usage of these APIs frequently leads to severe vulnerabilities such as remote code execution. Existing methods for generating API security rules are limited as they rely on incomplete documentation or infer patterns from source code based on discovered inconsistencies. We introduce VULGENIE, a patch-driven framework that extracts precise API security rules from confirmed security patches to then detect API misuse vulnerabilities. VULGENIE addresses three key challenges. First, it isolates violated constraints and defenses-related changes from noisy patches using our novel modification behavior dependency patch graph datastructure. Second, it identifies protected security-sensitive APIs and synthesizes rules through attack-defense crossvalidation. Third, it scales analysis with adaptive, deviationguided static analysis to balance precision and performance. Evaluated on 150 recent Java security patches, VULGENIE extracts 198 correct API security rules with 81.82% precision, uncovering 177 rules absent in CodeQL. On ten popular Java applications, VULGENIE detects 46 0-day vulnerabilities, substantially outperforming state-of-the-art works. Through our responsible vulnerability disclosure, 26 vulnerabilities have already been fixed with ten CVE identifiers assigned.
Autonomy Comes with Costs: Detecting Denial-of-Service Vulnerabilities Caused by Resource Abusing in LLM-based Agents
Authors:
Jiaqi Luo, Jiarun Dai, Fengyu Liu, Songyang Peng, Youkun Shi, Tong Bu, Geng Hong, Xudong Pan, and Yuan Zhang
Abstract:
LLM-based agents have recently attracted significant attention. By leveraging the semantic understanding capabilities of large language models (LLMs), these agents can autonomously perform complex tasks according to user requests, such as downloading files and summarizing content. However, the lack of comprehensive resource governance renders them susceptible to abuse, potentially leading to resource exhaustion and denial-of-service (DoS) conditions. In this work, we present the first systematic security study of resource management in LLM-based agents. We identify three representative patterns of resource lifecycle management, each of which enables distinct avenues for DoS exploitation. Building on these insights, we propose AgentDoS, a novel directed grey-box fuzzing framework designed to detect DoS vulnerabilities arising from resource exhaustion. AgentDoS first analyzes the resource lifecycle within the agent and then leverages an LLM to generate functionality-specific seed prompts in natural language that drive the agent toward excessive resource consumption. We evaluated AgentDoS on 20 widely used open-source LLM-based agents and discovered 36 zero-day vulnerabilities affecting 16 agents, 15 of which have over 10,000 stars on GitHub. To date, 15 CVE IDs have been assigned for these vulnerabilities.
Wormholes in the File System: Understanding the Misunderstanding of Symlinks
Authors:
Yongheng Liu, Lei Zhang, Yuhang Zhao, and Yuzhou He
Abstract:
Symlinks, a file system feature supported at the operating system level for more than four decades, are deeply integrated across the software stack, making them an indispensable component of modern computing environments. The symlink mechanism is highly flexible, allowing references to nearly any location in the file system and enabling complex resolution behaviors. However, this flexibility also introduces significant misunderstandings for developers, which has led to widespread misuse and become a primary vector for critical security vulnerabilities in file-handling software, e.g., arbitrary file write and code execution.
In this paper, we present the first systematic study of how symlinks are incorrectly handled and the substantial security risks. Specifically, we study how programming language standard libraries improperly utilize symlink-related system calls, and how such misuse and misunderstanding propagates to downstream applications which results in various inadequate path validations. To further understand whether these inadequate path validations could be exploited in real world, we design and implement five path validation bypass schemes and three arbitrary write methods. Our evaluation across 85 projects uncovered 16 arbitrary file write vulnerabilities, two of which have been assigned high-severity CVEs. We also demonstrate their severe impacts in real-world scenarios and propose multiple mitigation strategies. We have responsibly disclosed these issues to the affected vendors.
When Fun Turns Toxic: A First Look at Aggressive Advertising in Mini-games
Authors:
Pei Chen, Geng Hong, Yicheng Qin, Huazhe Wang, Mengying Wu, Min Yang, Ziru Zhao, Yuanpeng Zhu, and Tao Su
Abstract:
Mini-games have emerged as a dominant paradigm within super-app ecosystems, enabling lightweight services like casual games to reach millions of users instantly. While official advertisement interfaces simplify monetization, the ease of integration and insufficient oversight have led to aggressive and potentially deceptive advertising practices, severely degrading the user experience. Aggressive advertising, though not malware, still subverts platform security boundaries by abusing legitimate APIs to bypass auditing, manipulate user interaction, and undermine platform trust, constituting a systemic security risk rather than mere policy violation. In this work, we conduct the first systematic security analysis of aggressive advertising in mini-games. We analyze platform policies and developer capabilities across nine minigame platforms, and characterize aggressive advertising behaviors. We further design a scalable detection framework, MAAD, and perform a large-scale measurement across three major platforms, i.e., WeChat, Facebook Instant Games, and Quickgame, revealing that 49.95% of mini-games exhibit aggressive advertising, including cases in highly popular titles with over 100k user reviews. Our analysis further uncovers their disruptive behavioral patterns, such as game-specific triggers, excessive pop-up frequency, and misleading strategies, as well as adversarial bypass techniques. These findings uncover that aggressive advertising constitutes a widespread form of platform abuse enabled by structural blind spots in current enforcement mechanisms. We provide actionable implications for strengthening platform governance, detection, and long-term ecosystem resilience.
MATE: Policy-Aware Security Auditing for Mobile Agents via Synthesis-Driven Trajectory Learning
Authors:
Changyue Jiang, Jiayi Wang, Xin Wen, Jiarun Dai, Geng Hong, and Xudong Pan
Abstract:
LLM-based mobile agents are increasingly deployed to automate complex workflows in mobile applications, but their execution trajectories can violate application- and user-specific security policies. Existing trajectory-level safety approaches either rely on prompting general-purpose LLMs or on static pattern rules, making it difficult to enforce fine-grained natural-language policies and to generalize across heterogeneous apps, policies, and tasks. Moreover, training policy-aware security auditors is hindered by the lack of realistic, policy-conditioned mobile agent trajectories.
In this paper, we present MATE, a lightweight, policy-aware security auditing model for mobile agents. MATE jointly encodes execution trajectories and natural-language security policies to produce policy-conditioned violation judgments with explanations, treating security requirements as editable text rather than fixed model parameters. This design enables MATE to support application-specific and user-defined policies and to adapt to evolving security requirements without retraining. To train MATE, we extract functional descriptions, operation workflows, and security policy from 158 Chinese and English mobile applications and build a knowledge-grounded synthesis pipeline that generates semantically realistic, policy-conditioned agent trajectories with multi-stage quality control, yielding over 140K trajectories for training. Using this pipeline, we construct MATEBench, a trajectory-level security auditing benchmark with two synthetic subsets and one real-world subset of manually collected mobile agent trajectories. Models trained via synthesis-driven trajectory learning achieve over 95% auditing accuracy on MATEBench and over 90% accuracy on two external agent-security benchmarks, exceeding baselines by about 20% on average. Crucially, on the real-world subset of MATEBench, our best model also exceeds 95% accuracy, demonstrating that knowledge-grounded synthesis enables accurate and customizable security auditing for mobile agents under both application-derived and user-specified policies.