The discourse around Artificial Intelligence (AI) simultaneously claims this ballooning technology to be both a harbinger of doom and an innovative boon. Proponents have increasingly cited AI as a potential driver of efficiency and analytical neutrality, encouraging its application in new areas of work and study. Crucially, AI has come to be seen as a site of evolution for the legal profession and processes, with firms investing in legal research technologies to streamline operations and courts exploring the potential for the application of AI in sentencing decisions. But even more pertinent are the aspirations present amongst AI optimists for a larger and more potent role for AI in the future, be that in providing accessible legal advice to large groups of people or in analysing large quantities of legal data to help secure optimal outcomes for those using these new programmes. However, it is the position of this paper that these aspirations are, at best, utopian (failing to reconcile certain practical limitations of the technology with the legal profession and practice that it seeks to represent) and, at worst, dystopian (exacerbating existing violations of due process, undermining accessibility of justice, and reducing judicial legitimacy). As such, legal AI must be rigorously examined in relation to the validity of its inputs, processes, and backers in order to prevent pernicious path dependencies from rearing their heads in the years to come.