Hardening networks against strategic attackers using attack graph games

Abstract

We consider the problem faced by a network administrator (defender) when deploying limited security resources to protect a network against a strategic attacker. To evaluate the effectiveness of a defense strategy, one must consider possible counterattacks that an attacker can choose. We use game theory to model the interaction between the defender and the attacker. Game theory provides relevant concepts and algorithms for computing optimal strategies in environments with multiple decision makers. To model the space of attacker’s possible actions, we use attack graphs, that compactly represent all known sequences of attacker’s action that may lead to successful attack for a given network. We demonstrate our approach on a specific type of defense actions, where the defender deploys deceptive hosts and services (honeypots) to detect and mitigate attacks.We assume the worst-case attacker who has a complete knowledge of the (typically randomized) defense strategy. We seek the optimal defense strategy against this attacker in the form of a Stackelberg equilibrium. Computing this solution exactly using standard techniques has limited scalability, so we investigate several approaches for increasing scalability to realistic problems. We introduce optimization methods for finding exact solutions for these games and then propose a variety of polynomial heuristic algorithms that scale to significantly larger games. We analyze the scalability and the quality of these heuristic solutions on realistic network topologies. We show that the strategies found by the heuristics are often near-optimal and that they outperform non-game-theoretic baselines. Finally, we show how attack graph games can be used to answer various research questions relevant to network security administrators.