A team of researchers from Pennsylvania State University (PSU) and IBM have designed a proof-of-concept Android Trojan app that can steal passwords and other sensitive information by using the smartphone's motion sensors to determine what keys victims tap on their touchscreens when unlocking their phones or inputting credit card numbers during phone banking operations.
The Trojan horse is dubbed TapLogger by its creators and was designed to demonstrate how data from a smartphone's accelerometer and orientation sensors can be abused by applications with no special security permissions to compromise privacy.
TapLogger was created by Zhi Xu, a PhD candidate in the Department of Computer Science and Engineering at PSU, Kun Bai, a researcher at IBM T.J. Watson Research Center and Sencun Zhu, an associate professor of Computer Science and Engineering at PSU's College of Engineering.
Accelerometer and orientation sensor data are not protected under Android's security model, and this means that they are exposed to any application, regardless of its permissions on the system, the research team said in a paper that was presented during the ACM Conference on Security and Privacy in Wireless and Mobile Networks on Tuesday.
The TapLogger application functions as an icon-matching game, but has several background components that capture and use data from the motion sensors to infer touchscreen-based user input.
When certain regions of the touchscreen are tapped during the normal phone operation, the device experiences subtle moves. For example, tapping somewhere on the right side of the touchscreen, will cause the phone to tilt slightly to the right.
These phone movements are picked up by the motion sensors and can then be analyzed to build patterns corresponding to specific tap events when performing certain actions, like when typing the screen unlock PIN or entering the credit card number during a phone call.
After installation, TapLogger runs in training mode and collects motion sensor data while the user plays the icon-matching game. This is necessary because tap-generated movements can be different for every phone and user.
After it has collected enough data, the Trojan app builds tap event patterns and starts using them to infer user input during targeted operations.
"While the applications relying on mobile sensing are booming, the security and privacy issues related to such applications are not well understood yet," the researchers said in their paper, noting that other motion sensor-based attacks have been demonstrated in the past.
In August 2011, a pair of researchers from University of California proposed a similar attack and designed a concept application called TouchLogger to demonstrate it.
However, compared to TouchLogger, TapLogger uses additional orientation sensor readings and introduces the training mode for device-specific data. It also features stealth options and supports two practical attacks -- inferring screen unlock passwords and credit card PIN numbers, the new Trojan's creators said.
Another motion-sensor-based attack was presented in October 2011 by a research team from the Georgia Institute of Technology, who used data from an iPhone 4's accelerometer and gyroscope to infer what was being typed on a computer keyboard positioned near the device.