Probably to lock out future competition.

Note that Webtext and OpenWebText datasets were basically scraped from Reddit. It's incredibly valuable if you can lock down this dataset vs. letting competitors use it.