The main argument developed in this chapter is that AI tools may help governments in exercising executive power responsibly, but that this requires democratic conditions and highly cultivated administrators. It is argued that the current discourse that public bureaucracies ought to digitally transform their operations rests on a misconception of the nature of the administrative bureaucracy. A corresponding disregard of the responsibility inherent to the exercise of executive power follows from such misconception. Indeed, the expectation that government operations will be automatically enhanced through their incorporation of AI-tools is based on the assumption that public bureaucracies are procedure-governed machines, in which democratic values, human decisions, the quality of the mind, and power constellations do not matter. Embedding AI in government operations to achieve higher efficiency, speed and uniformity of operations only seems a logical step if civil service bureaucracies are conceived as machines. The Dutch childcare benefits scandal (2013–2019) and its aftermath (2019–2024) are discussed to illustrate the culture of irresponsibility that resulted from the integration of AI-fraud detection by the Dutch Tax Office. Such bureaucratic nightmare, it is concluded, cannot be prevented by simply integrating “explainable”, “responsible” or “human-centric” AI. Instead, it is argued that however accurate and “ethical” AI might become, it can never be given priority in decision-making regarding matters that have a higher political and cultural priority (like democracy) than efficient operations. This also means that AI-regulations must practically reinforce the consciousness that the fate of people’s lives lies in administrators’ hands.