CSAM is a canary for general AI safety. If we can’t prevent SD from creating CP, will we be able to stop robots from killing people?