Not sure, I assume someone reports it then it gets turned over to AI to convert voice->script and compare it to a large set of known antivax rhetoric, if it surpasses a threshold it gets banned and stays that way until the owner asks for a review and then at that point a human probably enters the chat.