Sort of what I do, posted in related topics here somewhere.
An AI generated/processed post always has hallmarks that give it away.
We parse new topic posts from low-level trust users through an agent which flags it as possibly AI or not, the user gets a polite message from the bot explaining why and suggesting what they should do about it, while we get to either override the flag or bin the post.
It works, but the reality is a wave of people who just cut and paste a conversation into the forum, which is not wanted.
It just results in a wall of words being summarised by an AI and humans becoming proxies.
It’s better to have a zero-tolerance policy, an exception being language issues ,but with auto-translation that isn’t necessary any longer. Policies also need to be visible, if this moderation goes on behind the scenes then you lose awareness.