Bushings are one of the primary causes of failures in power transformers, and that's why several offline and online predictive maintenance techniques have been developed to evaluate the state and condition of bushings. This paper will demonstrate the capability of an Autoencoder network to act as an anomaly detector, indicating in real time failures in high voltage capacitive bushings, using statistical parameters of the leakage current vectors as input for the model. The results obtained by this study show that anomaly detection techniques are promising for the online diagnosis of condensing bushings.