Evolutionary Generation of Diversity in Embedded Binary Executables for Cyber Resiliency

Hardening avionics systems against cyber attack is difficult and expensive. Attackers benefit from a break one, break all advantage due to the dominant mono-culture of automated systems. Also, undecidability of behavioral equivalence for arbitrary algorithms prevents the provable absence of undesired behaviors within the original specification. This research presents results of computational experiments using bio-inspired genetic programming to generate diverse implementations of executable software and thereby disrupt the mono-culture. Diversity is measured using the SSDeep context triggered piecewise hashing algorithm. Experiments are divided into two phases. Phase I explores the use of semantically-equivalent alterations that retain the specified behavior of the starting program while diversifying the implementation. Results show efficacy against tailored exploits. Phase II relaxes requirements on search operators at the cost of requiring functionality tests. Results show success in demonstrating the removal of undesired specified behaviors