Are You Spying on Me? Large-Scale Analysis on IoT Data Exposure through Companion Apps

Recent research has highlighted privacy as a primary concern for IoT device users. However, due to the challenges in conducting a large-scale study to analyze thousands of devices, there has been less study on how pervasive unauthorized data exposure has actually become on today's IoT devices and the privacy implications of such exposure. To fill this gap, we leverage the observation that most IoT devices on the market today use their companion mobile apps as an intermediary to process, label and transmit the data they collect. As a result, the semantic information carried by these apps can be recovered and analyzed automatically to track the collection and sharing of IoT data. In this paper, we report the first of such a study, based upon a new framework IoTProfiler, which statically analyzes a large number of companion apps to infer and track the data collected by their IoT devices. Our approach utilizes machine learning to detect the code snippet in a companion app that handles IoT data and further recovers the semantics of the data from the snippet to evaluate whether their exposure has been properly communicated to the user. By running IoTProfiler on 6,208 companion apps, our research has led to the discovery of 1,973 apps that expose user data without proper disclosure, covering IoT devices from at least 1,559 unique vendors. Our findings include highly sensitive information, such as health status and home address, and the pervasiveness of unauthorized sharing of the data to third parties, including those in different countries. Our findings highlight the urgent need to regulate today's IoT industry to protect user privacy.