Adversarial attacks and mitigation for anomaly detectors of cyber-physical systems

•Successfully evaded RNN-based anomaly detectors in real-world CPSs using a gradient-based adversarial attack.•Simultaneously evaded built-in rule checkers by optimising the attack with a genetic algorithm.•Optimised the adversarial attack for both continuous and discrete data.•Proposed a formal definition of CPS attacks in an adversarial setting.•Demonstrated the difficulty of re-training the anomaly detectors to catch these attacks.