The problem isn’t training data, it’s reference locking and allowing anyone to make whatever content they want.