perhaps they don't want the datasets that they use to train the AI to be watered down by other AI generated content. after all alot of the text data was sourced from the internet.