Analysis of Adversarial Learning of Reactor State

With the recent successful attack attempts against the digital control systems of critical infrastructures, there is a need to develop new defense strategies that take into account two important realities: state-sponsored attackers can rely on a number of techniques including espionage, social engineering, and brute force techniques, etc. to gain access to the raw data used to control system behavior; and attackers can falsify operational data in manners that do not trigger conventional outlier/anomaly detection techniques in order to remain undetected, which is referred to as false data injection attacks. This paper explores the use of model-based techniques which have been recently promoted as potential approach for identifying data injection attacks. Specifically, we explore whether attackers can emulate the predictions of the models used by the defender. For demonstration, we employ a simplified point kinetics models with a number of unknown parameters and explore how inference techniques may be used to fully determine the dynamical behavior of the system. Results indicate that attackers can emulate the model predictions with high accuracy, indicating that the brute force application of model-based defenses is not effective and must be supplemented by other defense measures.