EUVIMEDEUVIMED entdecken
Crossref

MrM: Black-Box Membership Inference Attacks Against Multimodal RAG Systems

Peiru Yang, Jinhua Yin, Haoran Zheng, Xueying Bai, Huili Wang, Yufei Sun, Xintian Li, Songwei Pei, Yongfeng Huang, Tao Qi

Proceedings of the AAAI Conference on Artificial Intelligence · 14.03.2026 · 40 (40) · 34295-34303

DOI: 10.1609/aaai.v40i40.40726 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence, online erschienen am 14.03.2026. Verfasst von Peiru Yang und 9 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Multimodal retrieval-augmented generation (RAG) systems enhance large vision-language models by integrating cross-modal knowledge, enabling their increasing adoption across real-world multimodal tasks. These knowledge databases may contain sensitive information that requires privacy protection. However, multimodal RAG systems inherently grant external users indirect access to such data, making them potentially vulnerable to privacy attacks, particularly membership inference attacks (MIAs). Existing MIA methods targeting RAG systems predominantly focus on the textual modality, while the visual modality remains relatively underexplored. To bridge this gap, we propose MrM, the first black-box MIA framework targeted at multimodal RAG systems. It utilizes a multi-object data perturbation framework constrained by counterfactual attacks, which can concurrently induce the RAG systems to retrieve the target data and generate information that leaks the membership information. Our method first employs an object-aware data perturbation method to constrain the perturbation to key semantics and ensure successful retrieval. Building on this, we design a counterfact-informed mask selection strategy to prioritize the most informative masked regions, aiming to eliminate the interference of model self-knowledge and amplify attack efficacy. Finally, we perform statistical membership inference by modeling query trials to extract features that reflect the reconstruction of masked semantics from response patterns. Experiments on two visual datasets and eight mainstream commercial visual-language models (e.g., GPT-4o, Gemini-2) demonstrate that MrM achieves consistently strong performance across both sample-level and set-level evaluations, and remains robust under adaptive defenses.

Zitieren

APA

Yang, P., Yin, J., Zheng, H., Bai, X., Wang, H., Sun, Y., Li, X., Pei, S., Huang, Y., & Qi, T. (2026). MrM: Black-Box Membership Inference Attacks Against Multimodal RAG Systems. Proceedings of the AAAI Conference on Artificial Intelligence, 40(40), 34295-34303. https://doi.org/10.1609/aaai.v40i40.40726

Vancouver

Yang P, Yin J, Zheng H, Bai X, Wang H, Sun Y, et al. MrM: Black-Box Membership Inference Attacks Against Multimodal RAG Systems. Proceedings of the AAAI Conference on Artificial Intelligence. 2026;40(40):34295-34303. doi:10.1609/aaai.v40i40.40726

BibTeX
@article{Yang2026,
  author = {Yang, Peiru and Yin, Jinhua and Zheng, Haoran and Bai, Xueying and Wang, Huili and Sun, Yufei and Li, Xintian and Pei, Songwei and Huang, Yongfeng and Qi, Tao},
  title = {MrM: Black-Box Membership Inference Attacks Against Multimodal RAG Systems},
  journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
  year = {2026},
  volume = {40},
  number = {40},
  pages = {34295--34303},
  doi = {10.1609/aaai.v40i40.40726}
}
Kostenlos starten →