Identifying attacks is key to ensure security in cyber-physical systems. In this note we remark upon the computational complexity of the attack identification problem by showing how conventional approximation techniques may fail to identify attacks. Then, we propose decentralized and distributed monitors for attack identification with performance guarantees and low computational complexity. The proposed monitors rely on the geometric framework proposed in [1], yet require only local knowledge of the system dynamics and parameters. We exploit a divide-and-conquer approach, where first the system is partitioned into disjoint regions, then corrupted regions are identified via distributed computation, and finally corrupted components are isolated within regions.