Observation-assisted heuristic synthesis of covert attackers against unknown supervisors

Synthesis of Maximally Permissive Covert Attackers Against Unknown Supervisors by Using Observations

“…Recently, we have considered a more practical setup where the model of the supervisor is not available to the adversary [24]. To compensate the lack of knowledge on the model of the supervisor, it is assumed in [24] that the adversary has recorded a (prefix-closed) finite set of observations of the runs of the closed-loop system. In this more challenging setup, a covert attacker needs to be synthesized based solely on the model of the plant and the given set of observations.…”

“…The difficulty of this synthesis problem lies in the fact that there can be in general an infinite number of supervisors which are consistent with the observations, rendering the synthesis approaches developed in the existing works ineffective. In [24], we have proposed a technique to compute covert damage-reachable attackers by formulating it as an instance of the supervisor synthesis problem on certain surrogate plant model, which is constructed without using the model of the supervisor. Due to the overapproximation in the surrogate plant model, the synthesized attacker in [24] cannot ensure the maximal permissiveness in general.…”

“…In [24], we have proposed a technique to compute covert damage-reachable attackers by formulating it as an instance of the supervisor synthesis problem on certain surrogate plant model, which is constructed without using the model of the supervisor. Due to the overapproximation in the surrogate plant model, the synthesized attacker in [24] cannot ensure the maximal permissiveness in general. It is worth noting that there is a gap between the de facto maximal permissiveness, assuming the model of the supervisor is known, and the maximal permissiveness (from the adversary's point of view) that could be attained with a limited knowledge of the model of the supervisor.…”

“…In this paper, as in [24], we also assume the model of the supervisor is not available to the adversary and the adversary can use the observations to assist the synthesis of covert attackers. We consider attackers whose attack mechanisms are restricted to sensor replacement/deletion attacks and actuator enablement/disablement attacks in this work.…”

“…We allow sensor replacement/deletion attacks 2 and actuator enablement/disablement attacks. In comparison, there are two limitations regarding the approach proposed in [24]: 1) it only provides a sound, but generally incomplete, heuristic algorithm for the synthesis of covert damage-reachable attackers due to the use of over-approximation in the surrogate plant, and 2) it cannot deal with actuator enablement attacks.…”