If I was designing an anti troll/bot system I'd implement a few things. Let's call any bad actor on here a bot/troll or broll for ease.
Reputation based posting isn't a bad idea if carefully done
When a broll is banned any users from the same ip are flagged a suspect and a subsequent ban causes delayed posting from that ip. If a VPN the instance host list has the same effect applied. Exceptions based on Reputation.
You can check if text is ai/llm generated and an automated api check before posting and immediate ban if found.
Checks on if a user posts inhumanly fast or is oddly active etc would be sensible.
Any broll system has to be adaptive and measures taken need to be kept secret (this post for example).