Neural network integrity validation

A method for securing the integrity of a neural network in a secure computer system, the method comprising:

• training the neural network to match digital samples to categories in a set of categories;
• training the neural network, when presented with at least one golden sample, which is a sample outside the set of categories, to output a probability vector indicative of a preposterous result that the golden sample is matched into a predefined category in the set of categories to which it does not belong;
• programing the secure computer system with the trained neural network;
• adapting the computer system to receive digital samples and to present the digital samples to the trained neural network;
• presenting, by the computer system, the golden sample to the trained neural network;
• and
• if the neural network outputs a probability vector classifying the golden sample into a predefined category to which it does not belong in a way that is a preposterous result, declaring the neural network as uncompromised and, otherwise, declaring the neural network as compromised.