That's just not an issue in practice. All major AI labs match generated code against their training data to prevent outputting verbatim copies. They're so confident they won't output copyrighted code that they offer copyright indemnification to paid users.