It may be a bit of a paradox, the idea of transparency in modelling while accessing large amounts of sometimes sensitive data - so what guard rails can AI have in place to ensure both transparency and functionality?