@Suiseiseki @lamp What I meant is, such models can also be trained just from "harmless" pictures of minors, and "adult" porn. If one really needs that kinda stuff. In that case no real person has been harmed.

If a model has been trained using real CSAI, then yes, the entire model should of course still be illegal to distribute with massive consequences, as doing so still perpetuated harm to actual children depicted in training data. As for pictures generated by said models, it is more complicated, as it ultimately depends on whether or not a real person can be recognized (and yes, this raises to the level where someone thought they "just used AI" and had no intention of harming a real child, but the model was massively overfit and always generated the same child).

But yeah, waste of investigators' resources is also an issue, but should not be a reason to ban something. Get gud at investigating, I guess.