Comment by giantg2

This raises an interesting point. Do you need to train models using CSAM so that the model can self-enforce restrictions on CSAM? If so, I wonder what moral/ethical questions this brings up.