

My guess is that the algorithm is really good at predicting who will be likely to follow that kind of content, rather than report it. Basically, it flies under the radar purely because the only people who see it are the ones who have a vested interest in it flying under the radar.
Aside from the other arguments people have presented, this wrecks one of the largest reasons that people produce CSAM. Pedophiles are insular data hoarders by necessity, because actually creating and procuring it is such a big risk. Every time they go online to find new content, they’re at risk of stumbling into a honeypot. And producing it requires IRL work, and a LOT of risk of being caught/turned in by the victim. They tend to form tight-knit rings, and one of the only reliable ways to get into a ring as an outsider is to provide your own CSAM to the others. CSAM is traded in these rings like baseball cards, where you need fresh content in order to receive fresh content.
The data hoarding side of things is where all of the “cops bust pedophile with 100TB of CSAM” headlines come from; In reality, it was probably like 1TB of videos, (which is a lot, but not unheard of) but was backed up multiple times in multiple places, because losing it would be catastrophic for the CSAM producer; They can’t simply go grab a new blue ray of it. And the cops counted the full size of each backup disk, not just the space that was used.
Intentionally marking your content as AI-generated would ruin the trading value, because nobody will see it as valuable/worth trading for if it’s fake. At best, you won’t get anything for it. At worst, you’d be labeled a cop trying to pass off AI content to gather evidence.