Beyond Visual Evidence Revealing and Mitigating Relational Privacy Leakage in Document MLLMs

Abstract

While the privacy risks of multimodal large language models (MLLMs) have drawn significant attention, the unique vulnerabilities of domain-specific MLLMs remain largely underexplored. Focusing on document understanding MLLMs for identity document processing, this paper investigates the privacy issues inherent in Key Information Extraction (KIE) tasks. We reveal that when input images lack sufficient visual evidence, these models often rely on memorized field relations from training data to infer missing content, thereby leaking multiple correlated fields containing sensitive personal information.

To mitigate this risk, we propose the Dynamic Relational Unlearning Framework (DRUF), comprising a Relational Decoupling Unlearning module and a dynamic set update mechanism. We also introduce DocPrivacyBench to evaluate privacy leakage under absent or minimal visual evidence. Across three MLLMs and six unlearning methods, DRUF improves leakage suppression by 4.8 percentage points over the strongest baseline while maintaining robust document information extraction performance.

BibTeX

@inproceedings{xu2026beyond,
  title     = {Beyond Visual Evidence: Revealing and Mitigating
               Relational Privacy Leakage in Document MLLMs},
  author    = {Xu, Beining and Wang, Hairui and Wang, Jiaxin and
               Chen, Changsheng and Chakraborty, Anirban},
  booktitle = {Proceedings of the 35th ACM International
               Conference on Multimedia},
  year      = {2026},
  note      = {Poster}
}