Artificial moral characters: constitutional AI and the challenge of alignment

Abstract This paper examines Constitutional AI through the lens of Aristotelian virtue ethics and 4E cognition in order to clarify how moral practice is transformed under conditions of AI-mediated action. It argues that while constitutionally trained language models exhibit stable, norm-guided behavioral patterns, these regularities do not constitute virtue or moral character in the Aristotelian sense. Virtue remains bound to embodied agency, affectivity, and practical judgment. The concept of Artificial Moral Character is therefore introduced as a heuristic tool for analyzing how human moral commitments are externalized and stabilized within technical systems through design, alignment, and governance. Drawing on 4E cognition, the paper develops an account of Extended Morality that shifts ethical attention from character attribution to the relational configuration of moral technological environments. Thereby, alignment is reframed as an ongoing practice of mediated responsibility rather than a property of artificial agents, highlighting the ethical significance of socio-technical infrastructures in shaping human judgment and accountability.