You here run into other problems though with people mirroring code on github that they do not have the rights to change the license for, which would be required for accepting the terms of services. E.g. if you take a GNU project hosted at Savannah and maintain your own independent for of it on github for whatever reason, you cannot re-license the bit that is own by the FSF to conform with the github ToS without getting the FSF's permission (and for other projects without a single copyright holder, every single contributor). This ofc isn't Microsoft's or Github's fault, but the people making the github repo, but still creates a problem for breaking licenses.

Their FAQ also has some ambiguity on what it has been trained on:

> It has been trained on natural language text and source code from publicly available sources, including code in public repositories on GitHub.