Back to AI News
arxiv

Specify Privacy Yourself: Assessing Inference-Time Personalized Privacy Preservation Ability of Large Vision-Language Models

The paper introduces Inference-Time Personalized Privacy Protection (ITP^3), a task in which users dynamically specify privacy boundaries through language. It presents SPY-Bench, described as the first systematic assessment of this ability, with 32,700 image-question samples and personalized privacy instructions across 67 categories and 24 real-world scenarios. Benchmarking 21 state-of-the-art LVLMs shows poor compliance, with even o4-mini reaching only about 24% accuracy, and limited contextual privacy understanding. The authors implement initial alignment methods, including a Noise Contrastive Alignment variant reporting 96.88% accuracy while maintaining reasonable general performance.

AI Brief

It provides a concrete benchmark and empirical evidence on a new personalized privacy-control task for vision-language models, useful for researchers working on LVLM safety and alignment.

arXiv:2611.13997v1 Announce Type: cross Abstract: Large Vision-Language Models (LVLMs) have demonstrated remarkable capabilities but raise significant privacy concerns due to their abilities to infer sensitive personal information from images with high precision. While current LVLMs are relatively well aligned to protect universal privacy, e.g., credit card data, we argue that privacy is inherently personalized and context-dependent. This work pivots towards a novel task: can LVLMs achieve Inference-Time Personalized Privacy Protection (ITP$^3$), allowing users to dynamically specify privacy boundaries through language specifications? To this end, we present SPY-Bench, the first systematic assessment of ITP$^3$ ability, which comprises (1) 32,700 unique samples with image-question pairs and personalized privacy instructions across 67 categories and 24 real-world scenarios, and (2) novel metrics grounded in user specifications and context awareness. Benchmarking the ITP$^3$ ability of 21 SOTA LVLMs, we reveal that: (i) most models, even the top-performing o4-mini, perform poorly, with only $\sim$24% compliance accuracy; (ii) they show quite limited contextual privacy understanding capability. Therefore, we implemented initial ITP$^3$ alignment methods, including a novel Noise Contrastive Alignment variant which achieves 96.88% accuracy while maintaining reasonable general performance. These results mark an initial step towards the ethical deployment of more controllable LVLMs. Code and data are available at https://github.com/achernarwang/specify-privacy-yourself
Be the first to rate this article.

Comments & Questions (0)

Captcha
Please be respectful — let's keep the conversation friendly.

No comments yet

Be the first to comment!