The thing that annoys me is that I could see value in AI moderation. Instead of scanning every post with AI with overly-broad criteria (and probably lower-power models), use AI to prescreen reports and determine whether they're worth surfacing to a human. IT could also be used to put temporary holds on material that's potentially criminal or just way over the line, but those holds should go to the very front of the human review queue to either be lifted or the content deleted.

Real moderation actions should not be taken without human input and should always be appealable, even if the appeal is just another mod looking at it to see if they agree.