Security Diffusion Games.

An increasing attention has been recently paid to studying the confrontation between a defender of a network and an aggressor who attacks selected targets and spreads the attack through the network in an infection-like manner. While to date the literature focused on the scenarios in which such spread follows a chance mechanism, in this paper we propose and analyze the first defender-attacker model in which the attacker is able to strategically guide the whole course of the attack spread. We find that even if the attacker knows the distribution of security resources chosen by the defender, finding an optimal plan of attack is NP-hard in a general case. However, we show that it is possible to efficiently find optimal strategies of the attacker and the defender for some simple network structures, such as cliques, stars, and trees. For more sophisticated network structures, we propose near-optimal strategies for both players.