Original title: AVP-Inspect: Coordinated Cyber-Physical Testing for Privacy Analysis of COTS Apple Vision Pro Applications
Yichang Xiong, Vamsi Shankar Simhadri, Yue Xiao, Xiaokuan Zhang · September 2026
Extended reality headsets collect huge amounts of personal data, which creates serious privacy risks. Researchers already studied Android-based headsets like Meta Quest, but Apple Vision Pro remains hard to examine because its software system is closed. To tackle this, researchers built an automated testing tool named AVP-Inspect to uncover privacy violations by monitoring outgoing internet traffic. The tool works by combining physical hardware controls that operate the headset with a special software engine that explores three-dimensional app interfaces. It also uses a standard system to categorize privacy rules and flag violations. The team tested their tool on a custom evaluation dataset, and then ran it on 324 real apps from the App Store for 20 minutes each. They found that 188 apps, or 58.0%, had at least one privacy violation. Furthermore, more than 60% of the network traffic flows sent out by these apps were not properly disclosed.
You hire a friend to clean your room, but they promise only to tidy the desk. You suspect they might also look through your drawers. You cannot watch them every second, so you set up a hidden camera. Here, the researchers built a mechanical hand to physically operate the headset and software to record all internet traffic, because a human tester would be too slow or miss secrets. The robot hand mimics a user, and the network monitor catches what data the apps actually send out.
If you own an Apple Vision Pro, apps may be sending data you never agreed to share. A new study found that 58% of tested apps had violations, and most network traffic is not properly disclosed. This means your eye movements, gestures, or surroundings could be leaked without your knowledge, undermining trust in privacy labels.
The study only analyzes network traffic, not other possible data leaks like local storage or code. Each app was tested for only 20 minutes, which may miss delayed or conditional behavior. The ground truth dataset was manually constructed, potentially introducing bias. Findings are specific to Apple Vision Pro and may not generalize to other XR devices or platforms.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
AVP-Inspect: Coordinated Cyber-Physical Testing for Privacy Analysis of COTS Apple Vision Pro Applications — Yichang Xiong, Vamsi Shankar Simhadri, Yue Xiao et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.