One might reasonably assume the creators of this stuf are training their models on real CSAM, and thus are guilty of many crimes in addition to the production. Everyone else training models to produce imagery uses real data, so why would AI generated CSAM be any different in this regard?