VR ProfiLens: User Profiling Risks in Consumer Virtual Reality Apps

TL;DR

VR ProfiLens introduces a top-down, law-grounded taxonomy and an empirical framework to quantify how much user attributes can be inferred from abstracted VR sensor data across consumer apps. By collecting multi-sensor streams (BM, EG, HJ, FE) from 20 participants across 10 popular VR apps and ground-truth surveys, the study demonstrates moderately high to high inference capability (up to $F1$ ≈ $90\%$) for a broad set of attributes, including demographics, anthropometrics, health, and interests. It shows that profiling risk varies by sensor type and app group, with multi-sensor adversaries amplifying exposure, and highlights design and regulatory implications such as sensor-level permissions, data minimization, obfuscation, and privacy-assistance tools. The work provides actionable guidance for developers and regulators to enhance transparency, user control, and privacy protections in VR environments, and proposes a framework that can be extended to evolving VR ecosystems. Overall, VR ProfiLens foregrounds substantial privacy risks in consumer VR and offers concrete avenues for mitigations and policy alignment.

Abstract

Virtual reality (VR) platforms and apps collect user sensor data, including motion, facial, eye, and hand data, in abstracted form. These data may expose users to unique privacy risks without their knowledge or meaningful awareness, yet the extent of these risks remains understudied. To address this gap, we propose VR ProfiLens, a framework to study user profiling based on VR sensor data and the resulting privacy risks across consumer VR apps. To systematically study this problem, we first develop a taxonomy rooted in the CCPA definition of personal information and expand it by sensor, app, and threat contexts to identify user attributes at risk. Then, we conduct a user study in which we collect VR sensor data from four sensor groups from real users interacting with 10 popular consumer VR apps, followed by a survey. We design and apply an analysis pipeline to demonstrate the feasibility of inferring user attributes using these data. Our results show that sensitive personal information can be inferred with moderately high to high risk (up to 90% F1 score) from abstracted sensor data. Through feature analysis, we further identify correlations among app groups and sensor groups in inferring user attributes. Our findings highlight risks to users, including privacy loss, tracking, targeted advertising, and safety threats. Finally, we discuss design implications and regulatory recommendations to enhance transparency and better protect users' privacy in VR.

Figures (5)

• Figure 1: Overview of VR ProfiLens. (1) Sensor Data Collection using BehaVRjarin2025behavr setup: Each user interacts with 10 consumer apps using Quest Pro while four sensor groups are recorded; Single-Sensor Adversaries have access to one sensor group, and Multi-Sensor Adversaries have access to multiple sensor groups; we grouped apps based on similarities of activities and emotional states. (2) Ground Truth Collection: Ground truth collection from other platforms or apps which is guided by VR User Profiling Taxonomy. Our taxonomy is rooted in law and expanded by threat scenarios and app groups as indicated by arrows. (3) Data Processing and Model Training: Sensor Data processing, feature engineering, and inference attack model training using sensor data and ground truth. (4) Profiling and Other Threats: An adversary can take only VR sensor data as model input to infer user attributes. Next, they initiate further attacks aligned with threat scenarios in Section \ref{['subsubsec:ThreatModel_ThreatScenarios']}.
• Figure 2: Feature Analysis on BM for Social and Archery App Groups. The Y-axis lists attributes and the X-axis shows top features. Color encodes feature importance: HI (high, pink), MH (medium-high, orange), MI (medium, yellow).
• Figure 3: Feature Analysis for BM Group across Different App Groups. Y-axis provides attribute names, X-axis represents corresponding top features for attribute inferences. Color code represents feature ranking: HI (high, pink), MH (medium-high, orange), MI (medium, yellow), while Circle size reflects feature frequency (i.e., larger circles, higher occurrences).
• Figure 4: Feature Analysis for FE Sensor Group Across Different App Groups. Y-axis provides attribute names, and X-axis represents corresponding top features for attribute inferences. Color code represents feature importance ranking: HI (high, pink), MH (medium-high, orange), and MI (medium, yellow), while circle size reflects feature frequency (i.e., larger circles indicate higher occurrences).
• Figure 5: Feature Analysis for HJ Sensor Group Across Different App Groups. Y-axis provides attribute names, and X-axis represents corresponding top features for attribute inferences. Color code represents feature importance ranking: HI (high, pink), MH (medium-high, orange), and MI (medium, yellow).