I agree that it's the ideal use case for the AI, it's just the issue there is that it's an area where you can't afford any false positives or negatives. All hell will break loose if sexual abuse content or footage of terrorism/cartel incidents ends up on random people's Facebook timelines, and another kind of hell will break loose if innocent people see their accounts nuked and their details reported to authorities because the system thought something they posted was illegal content.
We've already had both things happen in recent times, and it's been a legal and PR nightmare for affected companies. So while using AI would be great to avoid harm to any human moderators, the system would have to be perfect (or nearly perfect) in order to avoid any side effects that would cause serious harm to innocent people and the company/platform in question.