PrivAR: VLM-Based Privacy Risk Detection for Augmented Reality

Researchers have proposed PrivAR, a framework that uses vision language models (VLMs) with chain-of-thought prompting to detect context-dependent privacy risks in augmented reality (AR) environments. Unlike existing AR privacy frameworks that lack semantic understanding, PrivAR analyzes visual scene cues to infer sensitive information types—for example, identifying password notes in an office setting. It detects and obfuscates textual content to prevent exposure while preserving contextual cues needed for VLM inference. The system also explores contextually-informed warning interfaces to boost user privacy awareness. In experiments on a real-world AR dataset, PrivAR achieved 81.48% accuracy and an 84.62% F1-score. The work was published on arXiv (ID: 2604.22805).