{"ID":2838433,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2511.16940","arxiv_id":"2511.16940","title":"MultiPriv: Benchmarking Individual-Level Privacy Reasoning in Vision-Language Models","abstract":"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.","short_abstract":"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 privac...","url_abs":"https://arxiv.org/abs/2511.16940","url_pdf":"https://arxiv.org/pdf/2511.16940v3","authors":"[\"Xiongtao Sun\",\"Hui Li\",\"Jiaming Zhang\",\"Yujie Yang\",\"Kaili Liu\",\"Ruxin Feng\",\"Wen Jun Tan\",\"Wei Yang Bryan Lim\"]","published":"2025-11-21T04:33:11Z","proceeding":"cs.CV","tasks":"[\"cs.CV\",\"cs.CR\"]","methods":"[\"Language Model\",\"Generative Adversarial Network\"]","has_code":false,"code_links":[{"ID":606775,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_id":2838433,"paper_url":"https://arxiv.org/abs/2511.16940","paper_title":"MultiPriv: Benchmarking Individual-Level Privacy Reasoning in Vision-Language Models","repo_url":"https://github.com/CyberChangAn/MultiPriv-PII","is_official":false,"mentioned_in_paper":false,"mentioned_in_github":true,"github_stars":0}]}