Seems fishy… Can it do the reverse for proprietary code? If not, seems like it’s relying on being trained on the original code and not “clean room”.

That said, you fork it, you own it. Not technically a fork I guess but conceptually. And all code has bugs so welcome to your full time job maintaining 50 different previously freely maintained libraries.