There's almost certainly no way that LLMs are getting access to copywritten source code that's not publically available on github or any other public forum. Google is not hacking Microsoft for their source codes for training data. There's already a ton of training data freely available and accessible online. And the effort to payoff ratio of stealing uncompiled copywritten source code would be fucking massive, even before it get into the much higher legal consequences of corporate espionage and theft compared to copywrite violation.