Augmented Reality applications overlay our physical world with digital components in an interactive 3D space. These applications generally capture information about the physical world around the user through cameras and sensors, which can identify user movements and interactions with objects in the real world. In recent years, Location-Based Augmented Reality Games (LBARGs) have been used in several contexts, such as entertainment, tourism, and education. However, by capturing information about the environment, AR applications can lead to failures in maintaining user and bystander privacy. This paper addresses the identification and protection of sensitive data in LBARGs. We introduce LootAR, a location-based mobile AR game, and the SafeARUnity library, a real-time image processing middleware that acts as a layer between the AR application and the device’s camera, identifying and sanitizing sensitive data prior to rendering. Implementation aspects are discussed, involving Unity Sentis, a toolkit for running machine learning models in Unity, and YOLO, a fast single-stage object detector optimized for real-time applications. We also demonstrate the integration of SafeARUnity in mobile games, using LootAR as a case study.