To me, if you want to regulate controversial opinions, you have to err strongly to the side of too-open.
Remember, before the declaration of independence our founding fathers were terrorists/rebels. I don't mean this as a snappy hollow comparison. I'm saying fundamentally, you can't distinguish between a US soldier recruitment video and an ISIS soldier recruitment video without applying a moral context. How would an AI ever do this? And even if it could, who's moral retelling is the right one?
Better in my mind to stay out of the censorship game altogether and promote a forum that is inherently structure in a format that incentivizes accuracy over emotions.