There is only one way to regulate large language models-based AI: we have to regulate the choice of training data. Once all the data is mashed up it can't be untangled.