So just flood out the training data by putting enough counter narrative on the net to be scraped? I don't think that'd work. They'd build a model to tell them what political leanings each site had and run that before they approved each scrape for ingest into the training set.