Earlier quoted context omitted.
It is known that the LAION dataset underpinning foundation models like Stable Diffusion contained at least a few thousand instances of real-life CSAM at one point. I think you would be hard-pressed to prove that any model trained on internet scrapes definitively wasn't trained on any CSAM whatsoever. https://www.theverge.com/2023/12/20/24009418/generative-ai-i...
Then all image generation models should be considered inherently harmful, no?
So you'd think they would make youth services a good place to be for a child, right. After all, if that situation were to be only marginally better than child prostitution, there's no point to finding CSAM. Or at least, the point is not to protect children, since that is simply not what they're doing.
So how is youth services doing these days? Well ... NOT good. Regularly children run away from youth services to start doing child porn (ie. live off off an onlyfans account). There's a netflix series on the subject ("Young and locked up") which eventually, reluctantly shows the real problem, the outcome (ie. prison or street poverty).
In other words your argument doesn't really apply since the goal is not to improve children's well being. If that was the goal, these programs would do entirely different things.
Goals differ. There's people who go into government with the express purpose to "moralize" and arrest people for offenses. Obviously, to them it's the arresting part that's important, now how serious the offense was and CERTAINLY not if their actions actually help people. And then there's people who simply want a well-paying long-term job where they don't accomplish much. Ironically these are much less damaging, but they still seek to justify their own existence.
Both groups really, really, really want ALL image generation models to be considered inherently harmful, as you say.