Not only that, but with techniques like in-painting you could start with something that wasn't CP and then progressively make the model generate parts of it which then make up such an image right now. Stability saying they want to release an open version of SD that can't make CP is like a pen maker selling a pen that can't make CP: horseshit.
A human artist is obviously capable of generating CSAM, even if they have never seen that before.
Filtering of training data is countered by increasing capabilities to generalize:
Two years ago, that was a viable strategy: models could barely produce what was in the training data again.
Today models can generalize much better and compose concepts they have been trained on into new concepts that they haven’t.
Two years from now, filtering will be irrelevant.