mlsanityData to 7 Oct 2026

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

arXiv 2608.12911 ↗First posted 13 Aug 2026cs.CV cs.CR cs.MMcited by 0

On the map: Multimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness

also in 1 more
Abstract (arXiv)

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 make three key contributions.First, we propose the Dynamic Relational Unlearning Framework (DRUF) which comprises a Relational Decoupling Unlearning (RDU) module and a dynamic set update mechanism. It suppresses the leakage of high-risk field pairs while preserving KIE performance.Second, we introduce DocPrivacyBench, a novel benchmark to systematically evaluate a model's susceptibility to privacy leakage under conditions of absent or minimal visual evidence.Third, we evaluate three MLLMs and six unlearning methods using this benchmark, assessing both post-unlearning leakage suppression and utility preservation.Our results demonstrate that existing MLLMs consistently exhibit privacy leakage when visual evidence is scarce, particularly on noisier datasets. In contrast, DRUF outperforms the strongest baseline by improving leakage suppression by 4.8 percentage points, effectively mitigating privacy risks while maintaining robust document information extraction performance.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in MLLM jailbreaks and adversarial robustness, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 15 of its 27 references that are in the map. First posted on arXiv 13 Aug 2026.

Builds on

15 of 27 references.

Closest ancestors · 3

Same direction, newest first

  1. 2026Vision Language Model Helps Private Information De-Identification in Vision Data8 Jun
  2. 2026Unveiling Privacy Risks in Multi-modal Large Language Models: Task-specific Vulnerabilities and Mitigation Challenges8 Jun
  3. 2025A Survey of Safety on Large Vision-Language Models: Attacks, Defenses and Evaluations14 Feb

Other references · 12

  1. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 7,916
  2. 2020Extracting Training Data from Large Language Modelscited by 2,144
  3. 2022Knowledge Unlearning for Mitigating Privacy Risks in Language Modelscited by 424
  4. 2021Differentially Private Fine-tuning of Language Modelscited by 392
  5. 2021ICDAR2019 Competition on Scanned Receipt OCR and Information Extractioncited by 254
  6. 2021DocFormer: End-to-End Transformer for Document Understandingcited by 254
  7. 2021BROS: A Pre-trained Language Model Focusing on Text and Layout for Better Key Information Extraction from Documentscited by 104
  8. 2024Multimodal Unlearnable Examples: Protecting Data against Multimodal Contrastive Learningcited by 21
  9. 2024DocXPand-25k: a large and diverse benchmark dataset for identity documents analysiscited by 5
  10. 2024Unlearning Personal Data from a Single Imagecited by 4
  11. 2024Reconstructing training data from document understanding modelscited by 3
  12. 2023PrIeD-KIE: Towards Privacy Preserved Document Key Information Extractioncited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025On the Robustness of Vision-Language Models in Zero-shot Privacy Classification
  2. 2024The Phantom Menace: Unmasking Privacy Leakages in Vision-Language Models
  3. 2024Private Attribute Inference from Images with Vision-Language Models
  4. 2026ICU-Bench:Benchmarking Continual Unlearning in Multimodal Large Language Models
  5. 2026When Visual Privacy Protection Meets Multimodal Large Language Models
  6. 2026PPE-Bench: A Benchmark for Evaluating MLLM Unlearning under Private-Public Entanglement
  7. 2026Seeing Without Exposing: Adaptive Privacy Control for Open-World, Context-Hungry MLLMs
  8. 2025When Privacy Meets Recovery: The Overlooked Half of Surrogate-Driven Privacy Preservation for MLLM Editing