LLMs, by their very nature, don't have a concept of "source". Attribution is pretty much impossible. Attribution only really works if you use language models as "search engine". The moment you start generating output, the source is lost.