Discovering Application-Level Insider Attacks Using Symbolic Execution

This paper presents a technique to systematically discover insider attacks in applications. An attack model where the insider is in the same address space as the process and can corrupt arbitrary data is assumed. A formal technique based on symbolic execution and model-checking is developed to comprehensively enumerate all possible insider attacks corresponding to a given attack goal. The main advantage of the technique is that it operates directly on the program code in assembly language and no manual effort is necessary to translate the program into a formal model. We apply the technique to security-critical segments of the OpenSSH application.