Counter Autonomy Vulnerability Assessment

Counter Autonomy Vulnerability Assessment (CAVA) is designed to assess the countermeasures (CM) and counter-countermeasures (CCM) of autonomous agents, including scripted and AI-based agents. The system assumes that autonomy agents have defined tasks to complete, such as route planning or air-to-air combat in a multi-agent game.  CAVA includes an ontology that describes potential attack surfaces based on the development of the autonomy agents. Using this ontology approach, the system automatically explores attacks and counter-attacks on the autonomy agents. To assess the vulnerabilities of an autonomous agent, the CAVA system uses an adversarial attack suite, which consists of adversaries that are developed to identify and exploit vulnerabilities. The automated system attempts all known attacks on each agent to identify any weaknesses. The CAVA system also includes tools for analyzing the results of the assessment. These tools include evaluation metrics, replays, and the ability to analyze agent dynamics within an episode and across episodes. This allows users to understand how the agent behaves when it is attacked, and to identify any vulnerabilities that were exposed. Overall, the CAVA system is a comprehensive tool for assessing the vulnerabilities of autonomous agents.