Governing AI as infrastructure

Abstract Contemporary artificial intelligence (AI) governance frameworks predominantly view AI systems as tools or discrete instruments that augment human capabilities while remaining subordinate to human intention. This paper argues that as AI systems achieve infrastructural status—embedded, ubiquitous, and foundational to cognitive activity—and hence, they demand fundamentally different regulatory logics. From this lens, we develop a systematic framework that distinguishes tool properties (visibility, optionality, task-boundedness) from infrastructure properties (invisibility, obligatoriness, epistemic generativity). We then apply this framework to evaluate six AI governance instruments adopted in North America, Europe, and Asia. Our analysis reveals a systematic regulatory gap, wherein current frameworks address what AI systems do but not what they render thinkable. To bridge this gap, we introduce the concept of “epistemic infrastructure” to capture AI’s role in constituting the conditions of possibility for thought, and propose governance mechanisms calibrated to this infrastructural function.