[wkz]

If they really don't think that they need to comply with any license, then why not include all private repos in the training set? Could it be that they're worried about legal repercussions, whereas OSS is easier to (ab)use for this purpose because there's much less legal muscle behind it?

It is also very telling that they have not included any of their own proprietary code in the training set. If it's merely suggestions that are generated, why not also train on the NT kernel? Office?