In this paper, a covert attack detection scheme based on robust model control predictive arguments is developed for constrained cyber-physical systems. The core of the resulting solution relies on the combined use of receding horizon and set-theoretic control ideas that are exploited to generate redundant control sequences that are randomly selected on the actuator side so that the malicious agent is never aware on which control action has been indeed exploited. As a consequence, the attacker actions on the sensor-to-controller channel could lead to significantly erroneous data not complying with the expected evolution of the system modeling. Simulations on a B747 - 100/200 aircraft model show effectiveness and merits of the proposed approach.