The moat is the data they have collected. There is no way that the "de-identified" data from enterprise customers isn't making it into the training set one way or another.