Random thought of mine was companies use AI to moderate. But potentially malefactors can train the AI to flag harmless stuff. And because of the opaque nature of neural networks there isn't good mechanism to undo it, except by reverting.