Cyber Reactive Adversary Framework for Training (CRAFT)

Networks have become a critical battleground for military operations as adversaries become increasingly prolific and proficient at cyber warfare. While adversary threats have grown, training has largely remained focused on infrequent large-scale exercises that provide powerful training experiences, but are expensive and time-consuming. The DoD addresses this gap with the Persistent Cyber Training Environment (PCTE) to rapidly configure controlled virtual environments for training anytime, anywhere, but PCTE currently requires human experts to control adversary cyber operations forces (OPFOR). Providing PCTE with automated adversaries will both reduce the cost, and increase the frequency of training opportunities. Therefore, we propose to design and demonstrate the Cyber Reactive Adversary Framework for Training (CRAFT), a reactive agent framework and execution engine that inserts dynamic adversary behaviors into PCTE and provides assessment and authoring tools for instructors to understand and customize the training experience. CRAFT focuses on three components: (1) an Adaptive Adversary Framework that uses reactive behavior modeling to provide realistic and dynamic adversary behavior that is useful for training, (2) a Cyber Execution Engine that integrates adversary agents within the virtual network; and (3) an Instruction Support Suite that provides tools for configuring, assessing, adjusting, and revising adversary behaviors to provide effective training.