If Microsoft is so confident in the legality and ethics of Copilot, and that it doesn't leak or steal proprietary IP... they should go train it on the MS Word and Windows and Excel source trees.
Did they make a statement that they did not want to do that?
Because if not I would offer the very mundane explanation that the Copilot team probably just couldn't be bothered hitting up the other software teams and jumping through 3,046 internal red tape compliance steps to make their product 0.001% better (I am pretty sure the code base of all of GH dwarfs MS code base quite a lot)
I can't believe I am actually defending fucking Microsoft, but just want to say there isn't a conspiracy everwhere...
I have no doubt they will -- but the specific models will be used for Microsoft engineers. There will be a Copilot for Enterprise that trains on customers' private code.
What's that? They don't want to do that? Why not?