Whole Campaign Emulation with Reinforcement Learning for Cyber Test

Cyber-attacks pose existential, nation-level threats and directly challenge societal stability. The breadth of targets (small businesses to nation-states) and continuous nature of cyber-attacks make automated cyber test and evaluation (T&E) crucial to national security and domestic prosperity. Importantly, automation lowers the cost and increases the frequency of cyber T&E, thereby simultaneously increasing cyber test availability and coverage. Spurred by market demand as well as advancements in artificial intelligence (AI), automated approaches to penetration testing have seen a resurgence of interest in the academic literature. Yet to date, this burgeoning research community lacks a shared, long-term vision. Recently, we proposed a concept of whole campaign emulation (WCE) as a challenge problem and framework for automated penetration testing with reinforcement learning (RL) [1]. In this article, we review the state-of-the-art in RL-based automated penetration testing, assess its relation to WCE, and provide a case study using the open-source Network Attack Simulator (NASim) [2].