The real issue seems more about transparency and consent around how the models are trained and how author personas are being used