There's tooling out there 100% vibe coded, that is used by tens of thousands of devs daily, if that codebase found its way to training data, would it somehow ruin everything? I don't think this is really a problem, the problem will become people will need to identify good codebases from bad ones, if you point out which codes bad during training it makes a difference. There's a LOT of writings about how to write better code out there that I'm sure are already part of the training data.