Abstract: A computer-implemented method is disclosed for monitoring the logical consistency of an artificial neural network. Activation data of the artificial neural network, which are produced from input data, are initially read in. The activation data are transferred to at least one trained concept model which is trained to recognize and, if applicable, localize a partial feature of the features contained in the input data and to output a calibrated partial feature mask. The final output data are linked to the partial feature truth values by means of a fuzzy logic unit in such a way that a continual logical consistency truth value is produced therefrom. The logical consistency truth value is evaluated by means of an evaluation unit, wherein a logical inconsistency of the final output data is ascertained in an inconsistency region if the consistency truth value falls short of a predefined threshold.