arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2603.25403cs.CRcs.AIcs.LG
收藏

形状与物质:针对本地视觉-语言模型的双层侧信道攻击

Shape and Substance: Dual-Layer Side-Channel Attacks on Local Vision-Language Models

Ben-Gurion University of the Negev(内盖夫本-古里安大学)

Eyal Hadad, Mordechai Guri

AI总结:

本文针对本地视觉-语言模型的动态高分辨率预处理引入的算法侧信道,提出双层攻击框架,通过执行时间差异和缓存竞争分析,揭示隐私敏感内容的推理能力。

摘要:

在设备上的视觉-语言模型(VLMs)通过本地执行提供数据隐私。然而,我们展示了架构向动态高分辨率预处理(如AnyRes)的转变引入了固有的算法侧信道。与静态模型不同,动态预处理根据图像的宽高比分解图像为可变数量的块,产生依赖工作量的输入。我们展示了针对本地VLMs的双层攻击框架。在第一层,未授权攻击者可以利用标准未授权操作系统指标,通过执行时间差异可靠地指纹输入的几何形状。在第二层,通过分析最后级别缓存(LLC)的竞争,攻击者可以在相同几何形状内解决语义歧义,区分视觉密集(如医学X光)和稀疏(如文本文档)内容。通过评估最先进的模型如LLaVA-NeXT和Qwen2-VL,我们证明结合这些信号可以可靠地推断隐私敏感的上下文。最后,我们分析了缓解此漏洞的安全工程权衡,揭示了固定工作量填充带来的显著性能开销,并提出了安全边缘AI部署的实际设计建议。

英文摘要:

On-device Vision-Language Models (VLMs) promise data privacy via local execution. However, we show that the architectural shift toward Dynamic High-Resolution preprocessing (e.g., AnyRes) introduces an inherent algorithmic side-channel. Unlike static models, dynamic preprocessing decomposes images into a variable number of patches based on their aspect ratio, creating workload-dependent inputs. We demonstrate a dual-layer attack framework against local VLMs. In Tier 1, an unprivileged attacker can exploit significant execution-time variations using standard unprivileged OS metrics to reliably fingerprint the input's geometry. In Tier 2, by profiling Last-Level Cache (LLC) contention, the attacker can resolve semantic ambiguity within identical geometries, distinguishing between visually dense (e.g., medical X-rays) and sparse (e.g., text documents) content. By evaluating state-of-the-art models such as LLaVA-NeXT and Qwen2-VL, we show that combining these signals enables reliable inference of privacy-sensitive contexts. Finally, we analyze the security engineering trade-offs of mitigating this vulnerability, reveal substantial performance overhead with constant-work padding, and propose practical design recommendations for secure Edge AI deployments.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑