测试多模态检索增强生成系统是否泄露图像及描述信息
Do Multimodal RAG Systems Leak Data? A Comprehensive Evaluation of Membership Inference and Image Caption Retrieval Attacks
- 通过提示攻击检测mRAG中是否包含特定图像
- 能成功推断图像是否在数据库中并提取其标题信息
- 提醒需加强隐私保护，适合关注AI安全的研究者
多模态检索增强生成（mRAG）系统在视觉问答等任务中广泛应用，虽提升了性能，但也带来隐私风险。本文通过实证研究，考察标准提示下mRAG管道中的隐私漏洞。具体实施案例，试图判断某视觉资产（如图像）是否存在于mRAG中，并在存在时泄露其相关元数据（如标题）。结果表明，当前mRAG存在显著隐私泄露风险，亟需构建隐私保护机制，推动后续研究。代码已公开：https://github.com/aliwister/mrag-attack-eval。
原文摘要 · Abstract (English)
The growing adoption of multimodal Retrieval-Augmented Generation (mRAG) pipelines for vision-centric tasks (e.g., visual QA) introduces important privacy challenges. In particular, while mRAG provides a practical capability to connect private datasets and improve model performance, it risks the leakage of private information from these datasets. In this paper, we perform an empirical study to analyze the privacy risks inherent in the mRAG pipeline observed through standard model prompting. Specifically, we implement a case study that attempts to determine whether a visual asset (e.g., image) is included in the mRAG, and, if present, to leak the metadata (e.g., caption) related to it. Our findings highlight the need for privacy-preserving mechanisms and motivate future research on mRAG privacy. Our code is published online: https://github.com/aliwister/mrag-attack-eval.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。