Not sure what you mean by “look at the models”. You can see how they were trained. You could also look at their test procedures and results for the models. This is the fun part of AI models, give the model a novel condition and you have a probability of how it will behave. Hacking takes on a new and scarry meaning in an AI world.
The models can be perturbed by the training data used. So all of that will have to be examined.