Two Novel Defenses against Motion-Based Keystroke Inference Attacks
This addresses a security problem for smartphone users by mitigating privacy risks from sensor data without requiring app permissions, though it is incremental as it builds on known attack methods.
The paper tackles motion-based keystroke inference attacks on smartphones by proposing two defenses—reducing sensor data accuracy and random keyboard layout generation—which significantly reduce attack accuracy, as shown through implementation and user studies on Android.
Nowadays smartphones come embedded with multiple motion sensors, such as an accelerometer, a gyroscope and an orientation sensor. With these sensors, apps can gather more information and therefore provide end users with more functionality. However, these sensors also introduce the potential risk of leaking a user's private information because apps can access these sensors without requiring security permissions. By monitoring a device's motion, a malicious app may be able to infer sensitive information about the owner of the device. For example, related work has shown that sensitive information entered by a user on a device's touchscreen, such as numerical PINs or passwords, can be inferred from accelerometer and gyroscope data. In this paper, we study these motion-based keystroke inference attacks to determine what information they need to succeed. Based on this study, we propose two novel approaches to defend against keystroke inference attacks: 1) Reducing sensor data accuracy; 2) Random keyboard layout generation. We present the design and the implementation of these two defences on the Android platform and show how they significantly reduce the accuracy of keystroke inference attacks. We also conduct multiple user studies to evaluate the usability and feasibility of these two defences. Finally, we determine the impact of the defences on apps that have legitimate reasons to access motion sensors and show that the impact is negligible.