Well, given the “flaw” is a machine learning model, who was the developer and how do they relate to the training data?