BugInjector: Injecting Vulnerabilities for Configurable Cyber Defense

GrammaTech proposes Bug-Injector, a tool for generating cyber defense evaluation benchmarks through the injection of vulnerable code into existing host software. Bug-Injector is highly configurable, providing users the transparency required of a tool used to compare commercial products and perform security audits, and the customizability to enable focused evaluation of specific defensive tools, host programs, domains, and vulnerability classes. Bug-Injector is implemented independently of leading cyber defensive techniques, thus avoiding the potential circularity in which technical shortcomings limit the generated benchmark used to evaluate those very same techniques. Bug-Injector works by iteratively injecting vulnerabilities into a host program while simultaneously modifying, i.e. "mutating," the program to enable subsequent injections and camouflage previous injections. This process is guided using an evolutionary computation (EC) algorithm that favors candidate programs approximating a target vulnerability distribution and retaining syntactic and functional similarity to the original program.GrammaTech proposes Bug-Injector, a tool for generating cyber defense evaluation benchmarks through the injection of vulnerable code into existing host software. Bug-Injector is highly configurable, providing users the transparency required of a tool used to compare commercial products and perform security audits, and the customizability to enable focused evaluation of specific defensive tools, host programs, domains, and vulnerability classes. Bug-Injector is implemented independently of leading cyber defensive techniques, thus avoiding the potential circularity in which technical shortcomings limit the generated benchmark used to evaluate those very same techniques. Bug-Injector works by iteratively injecting vulnerabilities into a host program while simultaneously modifying, i.e. "mutating," the program to enable subsequent injections and camouflage previous injections. This process is guided using an evolutionary computation (EC) algorithm that favors candidate programs approximating a target vulnerability distribution and retaining syntactic and functional similarity to the original program.