PriMobiBench: Characterizing Visual Privacy Leakage in VLM-Driven Mobile GUI Agents

Qihang Cen , Tianshuo Cong , Da Song , Xinlei He , Jiaxing Song , Ke Xu , Qi Li

🏛 Institutions: Tsinghua , Shandong University , Wuhan University
📅 Date: September 12, 2026
📑 Publisher: arXiv
💻 Env: Mobile
🔑 Keywords: benchmark dataset privacy PriMobiBench MobiLeak user profiling

TLDR

PriMobiBench is a benchmark for visual privacy leakage in screenshot-driven mobile GUI agents, paired with MobiLeak, a dataset of execution traces from 16 apps covering 25 privacy attributes across 2,960 embedded privacy instances. VLMs extract sensitive on-screen information with up to 82.5% success and infer user profiles from aggregated visual evidence about 70% of the time; masking privacy-sensitive but task-irrelevant UI elements before cloud processing cuts profiling success by up to 58% at roughly 8% performance loss.

Open paper Report issue