This article examines the paradox of algorithmic legitimacy, defined as the contradiction between the growing trust in artificial intelligence for managerial decision-making and the erosion of accountability for the outcomes of such decisions. The study characterizes typical scenarios of managerial failures when utilizing algorithmic systems in commerce, finance, human resource management, logistics, and public administration. Furthermore, it uncovers the root causes of accountability erosion, specifically focusing on automation bias, the opacity of algorithmic models, and the lack of institutional control mechanisms. The author describes an «accountability architecture» model that integrates technical, institutional, legal, organizational, and educational levels of control. Additionally, a classification of artificial intelligence systems based on their level of autonomy is proposed, and corresponding control mechanisms are defined. The study concludes with the necessity of transitioning from declarative ethical principles to mechanisms of distributed control.
