Algorithmic accountability has emerged as a package of legal ideas that, on one hand, attempt to impose administrative law mechanisms such as transparency and due process on automated decision-making systems, and on the other hand, has developed computational approaches to constraining machine learning. In particular, by ensuring the complex computational analysis of individuals through machine learning models occurs more ‘fairly’, and is more explainable. As well as describing the necessity for computational legal implementations that actively constrain how data processing occurs, the chapter argues that there are risks that these mechanisms may involve ceding to data science and its corporate stakeholders the epistemological terrain as to what types of calculations are ‘fair’ and what type of information is an ‘explanation’.