Game-theoretic intrusion response and recovery

The severity and number of intrusions on computer networks are rapidly increasing. Preserving the availability and integrity of networked computing systems in the face of those fast-spreading intrusions requires advances not only in detection algorithms, but also in intrusion tolerance and automated response techniques. Additionally, the rapid size and complexity growth of computer networks, and their recently increasing integrations with physical systems signify the quest for systems that detect their own compromises and failures and automatically repair themselves. In particular, the ultimate goal of the intrusion tolerant system design is to adaptively react against malicious attacks in real-time, given offline knowledge about the network's topology, and online alerts and measurements from system-level sensors. Addressing all the practical and theoretical difficulties in design and deployment of an intrusion response framework in practice is a challenging problem. In particular, at each time instant, the response system needs to accurately determine the current security state of the system, given online sensory information. Moreover, decision upon a proactive strategy against attackers requires the knowledge about possible future attacks, or equivalently, system vulnerabilities and how to monitor and detect exploitations of those vulnerabilities. Additionally, prioritizing a specific response strategy over all other possible strategies demands an algorithm to compare criticality levels of compromised system assets. Finally, an efficient mathematical decision-making framework is needed to select the optimal response strategy by taking into account the possible future exploitations and damages as well as the criticality level of potentially compromised systems assets. This dissertation proposes a model-based solution to building a theoretically well-founded automated intrusion response and recovery framework in practice. In particular, we present an approach to address each of the abovementioned challenges. In particular, we introduce a security state estimation algorithm for cyber-physical networks that accounts for inherent uncertainties in intrusion detection systems' alert notifications and sensor measurements. To update the knowledge about the possible future attacks, our proposed intrusion forensics algorithm gradually identifies previously unknown system vulnerabilities and updates the deployed set of monitors every time a zero-day exploitations happens. Additionally, we introduce a consequence-centric security metric to automatically determine criticality level of the compromised system assets in any system state. Finally, to choose online countermeasure actions, we propose an intrusion response and recovery solution which employs a game-theoretic response strategy against adversaries by modeling the interaction between the system and the attacker as a two-player sequential game. The engine chooses optimal response actions by solving a partially observable competitive Markov decision process model. We validate the proposed solutions using real-world implementations, and showed that the proposed response system can efficiently recover the compromised computer networks automatically back to their normal operational mode