Issue is, AI is often trained on real children, sometimes even real CSAM(allegedly), which makes the “no real children were harmed” part not necessarily 100% true.

Also since AI can generate photorealistic imagery, it also muddies the water for the real thing.