The whole point of GANs is to train a model until it can defeat detection by another model. Virtually any technique for detection that can be reproduced could just be incorporated into the training loop and rendered almost useless.