Automated Asymmetric Red Teams (AART)

Red teaming is a technique that has been used successfully for some time in the military community to uncover system vulnerabilities or to find exploitable gaps in operational concepts, such as new weapon systems or tactics. Red teaming is currently a human intensive technique that typically brings together experts relevant to the system under consideration and who are then charged with identifying weaknesses. The primary challenge in this process is overcoming the limited exploration of possibilities and the potential bias of human experience and these limitations will likely hold for most adversary models that humans would build. A significant opportunity exists in automatically generating realistic adversary models that could potentially be used in any DoD simulation-based system, whether the goal of that simulation-based system is training, analysis, acquisition, or test and evaluation. This tool could also be used in homeland security and law enforcement, where adversaries in this case could be any agent out to do us harm. The overall objective of this effort is to research and design an asymmetric adversarial behavioral generation system (AART) that can provide computer-generated adversaries using asymmetric tactics and strategies in synthetic training environments. The following are specific technical objectives for this research.