Misskey already has support for computer vision models to auto-flag federated sexual content iirc.
I believe ROOST is trying to solve this issue but I don't hear anything about them on fedi, not even from admins who would benefit from easier moderation (on the ATProto side, Bluesky is a partner and ROOST set up shop there, some of Bluesky's projects are listed). Some of their stuff includes LLMs for trust & safety/moderation purposes, but IDK if there are LLMs actually being used in that dept atm tho.
I am all for automated moderation either way for as long as the big names in fedi want to compete against the bigger tech platforms instead of focusing on communities because the protocol is more suitable for that rather than Twitter 2. (Eugen in particular has an axe to grind against Bluesky and ATProto, which makes me believe that he wants Mastodon wants be the big social platform).