The deep (layered) state: shadow AI as statecraft dispositif

This article sociologically examines the emergence for ‘shadow AI’ within Canadian and Québecois public administrations. Drawing on 16 semi-structured interviews with federal and provincial bureaucrats, it analyzes unsanctioned or weakly disclosed uses of artificial intelligence (AI) as a statecraft dispositif that shapes governance and expertise distribution. The study combines constructivist political sociology and critical AI studies to show how AI adoption reorganizes power relations inside public adminsitration. Two contrasted case studies are developed. In Québec, the Ministère de la Cybersécurité et du Numérique embodies a centralized, “control-through-decree” oriented approach while at the federal level, decentralized and permissive model tolerates local experimentation and open-source tools. Across both cases, shadow AI appears as a constitutive feature of AI-powered statecraft. The article argues that AI’s black-boxed ontology, combined with distributed expertise, deepens layers of state secrecy. Ultimately, shadow AI contributes to an emergent ‘deep layered state’, characterized by experimentation, opacity, and adaptive governance.