Researchers at the University of Cambridge have developed a method for stealing a pin code for a smartphone using the device’s microphone and camera.
With a program called PIN Skimmer, the research team found they could identify the code by having the software use the camera to watch the user’s face and listen to the click of the buttons as they enter their PIN code. The attack assumes the victim is using a numbers-only soft keyboard.
“The microphone is used to detect touch events, while the camera is used to estimate the smartphone’s orientation, and correlate it to the position of the digit tapped by the user,” according to a paper written by the researchers. “We present the design, implementation and early evaluation of PIN Skimmer, which has a mobile application and a server component. The mobile application collects touch-event orientation patterns and later uses learnt patterns to infer PINs entered in a sensitive application.”
According to the researchers, using a test set of 50 4-digit PINs, PIN Skimmer can be used to infer more than 30 percent of PIN codes after two attempts. It also works on more than 50 percent of PINs after five attempts on Nexus S and Galaxy S3 phones running Google Android.
“When selecting from a set of 200 8-digit PINs, PIN Skimmer correctly infers about 45 percent of the PINs after 5 attempts and 60 percent after 10 attempts,” the paper states. “It turns out to be difficult to prevent such side-channel attacks, so we provide guidelines for developers to mitigate present and future side-channel attacks on PIN input.”
PIN Skimmer has four modes of operation: monitoring mode, collecting mode, learning mode and logging mode. Because the extraction of features, training of the learning algorithm and the predictions are performed on the server side, there is no noticeable effect on the battery when transferring data to a remote server, the researchers explained. Likewise, the image processing algorithms are run on a remote server and not on the phone for the same reason. The data uploaded for each mode of operation does not exceed 2.5 MB.
“The use of camera must also be stealthy in order not to raise the victim’s suspicion,” the researchers note. “Some phones have a LED that is automatically turned on when the camera is in use. The LED can be disabled via an API exposed by the Android OS. One possible drawback of this method is that it might not be supported by all android phones because of manufacturers’ customizations to the OS.”
In addition, some phones also make a shutter sound when pictures are taken. That can be disabled by muting the speakers muted while taking pictures.
“Designers must be aware of covert channel risks and engineer the overall system accordingly,” the paper concludes. “On smart OSes like Android, reasoning about the security of a trusted path becomes more complex as new features and services are added over time.”
