LAUSR

Several industry sectors, including critical infrastructure, have experienced severe cyber attacks against their Industrial Control Systems (ICS) due to the malware that masqueraded itself as a legitimate ICS process and communicated with valid ICS messages. Such behavior is difficult to detect by standard techniques. Intrusion Detection Systems (IDS) usually filter illegitimate communication using pre-defined patterns while statistical-based Anomaly Detection Systems (ADS) mostly observe selected attributes of transmitted packets without deeper analysis of ICS messages. We propose a new detection approach based on Deterministic Probabilistic Automata (DPAs) that capture the intended semantics of the ICS message exchange. The method models normal ICS message sequences using a set of DPAs representing expected traffic patterns. Then the detection system applies reasoning about the model to reveal a malicious activity in the ICS traffic expressed by unexpected ICS messages. In this paper, we significantly improve the performance of the automata-based detection method and reduce its false-positive rate. We also present a technique that produces additional details about detected anomalies, which is important for real-world deployment. The approach is demonstrated on IEC 104 or MMS communication from different ICS systems.