Now I’m just imagining a bunch of NLP models subtly learning large scale fraud.