A method for protection from cyber attacks in a communication network of a vehicle comprising: the steps of building sets of dominant voltage measurements for each message identifier associated to a message that is passing; extracting statistical features; supplying the statistical features for each message identifier that are available at each instant at input to a neural network of a pattern-recognition type; carrying out an operation of classification, or pattern recognition, supplying a prediction of a membership class corresponding to a given node on the basis of the statistical features supplied at input; evaluating whether the prediction supplied by the neural network corresponds to a given node that allows as admissible message identifier the message identifier at input and, if it does not, signalling an anomaly for the message identifier; and evaluating whether a number of anomalies signalled for said message identifier exceeds a given threshold.