That way filtering can be done on the client-side, and users aren't so dependent on the community admin to do the filtering. Not sure the final architecture. Forums are still highly centralized.
Cryptopanic.com is an interesting site with a baseline look and feel and comments integrated so something like that but running locally. Then an easy way to "mark as bot" button for training.
The part that bugs me most is they fill out fake 'About Me' sections on their profile.
There is no such thing as due process for AI agents. They are guilty until proven otherwise.