Abstract
Over the past decade, legal and philosophical debates have largely focused on whether autonomous artificial intelligence (AI) systems should be held criminally liable. While vital for advancing normative inquiries, this discourse, centered on responsibility gaps, legal personhood, and institutional design, has left unexplored a more fundamental question: whether intelligent systems can themselves exhibit criminal behavior, and why such behavior emerges. This article lays the groundwork for machine criminology, a new field of research dedicated to investigating the determinants of criminal conduct by intelligent systems. By shifting focus from attribution of liability to causation, machine criminology examines behavior, by trying, inter alia, (i) to define AI behavior that may qualify as “criminal;” (ii) to explain how the “decision” to engage in deviant behavior may be taken by machines; and (ii) to propose a preliminary etiology of machine crime, exploring how technical configurations, training environments, and interaction dynamics can foster criminogenic tendencies in machines, and why a residual propensity for deviance may be structurally unavoidable in complex systems. In doing so, machine criminology reframes the field’s epistemic priorities: from assigning liability to understanding causes, patterns, and propensities of machine deviance. This shift not only grounds normative debates in empirical analysis but also opens a forward-looking research agenda. By advancing machine criminology, this article calls for an anticipatory governance framework capable of safeguarding justice in an age where deviance may no longer be exclusively human.