paperMarch 2026Unreviewed

DP^2-VL: Private Photo Dataset Protection by Data Poisoning for Vision-Language Models

Hongyi Miao, Jun Jia, Xincheng Wang, Qianli Ma, Wei Sun, Wangqiu Zhou, Dandan Zhu, Yewen Cao, Zhi Liu, Guangtao Zhai

Abstract

Recent advances in visual-language alignment have endowed vision-language models (VLMs) with fine-grained image understanding capabilities. However, this progress also introduces new privacy risks. This paper first proposes a novel privacy threat model named identity-affiliation learning: an attacker fine-tunes a VLM using only a few private photos of a target individual, thereby embedding associations between the target facial identity and their private property and social relationships into th

Framework mappings

OWASP Top 10 for LLM Applications

LLM04Data and Model Poisoning

MITRE ATLAS

AML.T0020Poison Training Data

Suggested from the entry's categories.

Cite

@misc{miao2026dp2vl,
  title = {{DP\textasciicircum{}2-VL: Private Photo Dataset Protection by Data Poisoning for Vision-Language Models}},
  author = {Hongyi Miao and Jun Jia and Xincheng Wang and Qianli Ma and Wei Sun and Wangqiu Zhou and Dandan Zhu and Yewen Cao and Zhi Liu and Guangtao Zhai},
  year = {2026},
  month = mar,
  eprint = {2603.23925},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2603.23925}
}