Yeah this is bunch of fearmongering. Natural language processing, or to be more precise, sentiment analysis and contextualization, is just not ready to use, nor do I think they will be in the considerable future.

There will be way too many false-positives for them to be useful (i.e. "whypipo are the devils" will be removed and "gm frens" and cozyposting won't, which is opposite of what they want).

Overreach of this AI will be far too great and it can potentially wreck platforms by alienating the userbase (which is something I welcome).