Ironically though this makes the reverse a bit more defensible (i.e. using an LLM to reverse engineer a proprietary app) because that proprietary app's source code is less likely to be among the publicly available dataset.

But I imagine the corpos aren't going to look fondly on that for obvious reasons.