SafeScreen: Evaluating a Screen-Detecting Smartphone Camera App under Benign and Adversarial Use

DOI

Camera-equipped smartphones pose security risks to organisations by allowing intentional or accidental leaks of confidential on-screen information. We introduce SafeScreen, an Android camera app that detects and obfuscates screen content in real-time using deep-learning recognition for distant screens and Moiré pattern detection for close-up screen captures. Our mixed-methods, ecologically focused study compared "benign" (ordinary photography) and "malign" (circumventing detection) uses. Results show SafeScreen effectively prevents accidental leaks, but that the majority of users were able to exploit it by discovering workarounds such as partial screen occlusion. Our work contributes (1) a novel screen-blocking camera system, and (2) insights from real-world, unguided interactions. We show how evaluating security systems in authentic settings uncovers user-driven vulnerabilities and frustrations that inform future researchers and organisations. We close by discussing future technical features which could offer usability or security improvements, as well as emphasising the benefits of unscripted and adversarial user evaluations.