arXiv 2511.169402026-05-29cs.CVcs.CR

MultiPriv: 视觉语言模型中个体级隐私推理的基准测试

MultiPriv: Benchmarking Individual-Level Privacy Reasoning in Vision-Language Models

Xidian University（西安电子科技大学）；Nanyang Technological University（南洋理工大学）

Xiongtao Sun, Hui Li, Jiaming Zhang, Yujie Yang, Kaili Liu, Ruxin Feng, Wen Jun Tan, Wei Yang Bryan Lim

AI总结：

针对视觉语言模型通过层次化链式推理关联多模态数据识别个体的隐私风险，提出首个系统评估个体级隐私推理的基准MultiPriv，包含隐私感知与推理框架、双语多模态数据集和九项挑战任务，对50多个开源和商业模型评估发现60%的模型能以高达80%的准确率进行个体级隐私推理。

摘要：

现代视觉语言模型（VLM）通过层次化链式推理将碎片化的多模态数据与可识别的个体关联起来，构成了显著的个体级隐私风险。然而，现有的隐私基准在结构上不足以应对这一威胁，因为它们主要评估隐私感知，而未能解决更关键的隐私推理风险：VLM推断和关联分布式信息以构建个体档案的能力。为填补这一空白，我们提出了MultiPriv，这是第一个旨在系统评估VLM中个体级隐私推理的基准。我们引入了隐私感知与推理（PPR）框架，并构建了一个包含合成个体档案的双语多模态数据集，其中标识符（如人脸和姓名）与敏感属性相关联。该设计支持九项具有挑战性的任务，涵盖属性检测、跨图像重新识别和链式推理。我们对超过50个开源和商业VLM进行了大规模评估。在我们的受控基准中，60%的广泛使用的VLM能够以高达80%的准确率进行个体级隐私推理，这表明对个人隐私存在重大潜在威胁。该基准可在https://github.com/CyberChangAn/MultiPriv-PII获取。

英文摘要：

Modern Vision-Language Models (VLMs) pose significant individual-level privacy risks by linking fragmented multimodal data to identifiable individuals through hierarchical chain-of-thought reasoning. However, existing privacy benchmarks remain structurally insufficient for this threat, as they primarily evaluate privacy perception while failing to address the more critical risk of privacy reasoning: a VLM's ability to infer and link distributed information to construct individual profiles. To address this gap, we propose MultiPriv, the first benchmark designed to systematically evaluate individual-level privacy reasoning in VLMs. We introduce the Privacy Perception and Reasoning (PPR) framework and construct a bilingual multimodal dataset with synthetic individual profiles, where identifiers, such as faces and names, are linked to sensitive attributes. This design enables nine challenging tasks spanning attribute detection, cross-image re-identification, and chained inference. We conduct a large-scale evaluation of over 50 open-source and commercial VLMs. In our controlled benchmark, 60% of widely used VLMs can perform individual-level privacy reasoning with up to 80% accuracy, suggesting a significant potential threat to personal privacy. The benchmark is available at https://github.com/CyberChangAn/MultiPriv-PII.

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。