Optimal interdiction of attack plans

We present a Stackelberg game model of security in which the defender chooses a mitigation strategy that interdicts potential attack actions, and the attacker responds by computing an optimal attack plan that circumvents the deployed mitigations. First, we offer a general formulation for deterministic plan interdiction as a mixed-integer program, and use constraint generation to compute optimal solutions, leveraging state-of-the-art partial satisfaction planning techniques. We also present a greedy heuristic for this problem, and compare its performance with the optimal MILP-based approach. We then extend our framework to incorporate uncertainty about attacker’s capabilities, costs, goals, and action execution uncertainty, and show that these extensions retain the basic structure of the deterministic plan interdiction problem. Introduction of more general models of planning uncertainty require us to model the attacker’s problem as a general MDP, and demonstrate that the MDP interdiction problem can still be solved using the basic constraint generation framework.