An interface would have to be built that regulates content outputs based on training criteria, just like any other AI integration used to inform. I'm not sure what's complicated about that? Educational tools have been produced this way for a long time, but not with such expansive libraries behind it.