I see. But in order for that to operate it requires all written text to be processed through an algorithm. For it to understand "threats of violence", it has to understand context, not just count the words. So it essentially has to be a lesser version of the YouTube censorship algorithm.