Google Leaks PDF Download

As mobile devices become more widespread and powerful, they store more sensitive data, which include not only personal user information but also data collected via sensors on the device. When mobile applications have access to this sensitive information, they may leak it accidentally or by malicious design. Google's Android operating system provides a permissions-based security model that restricts an application's access to sensitive data. Each application statically declares the sensitive data and functionality that it requires in a manifest, which is presented to the user for approval during installation. However, it is difficult to determine how sensitive data will be used once the application has been installed. To address this problem, we present AndroidLeaks, a static analysis framework for automatically finding potential leaks of sensitive information in Android applications on a massive scale. AndroidLeaks leverages Android's permission scheme to identify sources of private data, then performs taint-aware slicing to determine if private data will be leaked via a network sink. We evaluated AndroidLeaks on 24,350 Android applications from several Android markets. AndroidLeaks found 57,299 potential privacy leaks in 7,414 Android applications, of which we have manually verified that 2,342 applications leak private data, including phone information, GPS location, Wi-Fi data, and audio recorded with the microphone. While previous work, such as TaintDroid, has effectively analyzed the data leakage of a small set of applications, no previous Android analysis tool has been able to effectively evaluate the leakage of a large set of applications in a reasonable amount of time. AndroidLeaks examined these applications in 30 hours, which indicates that it is capable of scaling to the rate at which new applications are developed. As ad code makes up a substantial percentage of the overall leaks that we discovered, we further investigated thirteen ad libraries. We discovered that ad libraries will frequently attempt to access sensitive content beyond that which is required to target ads, such as a user's contact book or calendar. Further, we identified four ad libraries that introduce a vulnerability to any application in which they are included. By exploiting this vulnerability, an attacker can instruct a user's device to perform various actions including placing a phone call, sending an SMS or email, and modifying contact and calendar entries. We propose solutions to problems caused by the lack of privilege separation between application code and ad code and discuss difficulties in addressing the vulnerabilities we discovered.