Governing Machines at the Edge of Failure: A Layered Responsible AI Framework for Critical Infrastructure Systems

Abstract

The accelerating integration of artificial intelligence into energy grids, water systems, transportation networks, and healthcare facilities has outpaced the maturity of governance instruments designed to keep these systems safe, explicable, and accountable. This paper addresses a persistent tension in the responsible artificial intelligence literature: high-level ethical principles have converged internationally, yet operational guidance for critical infrastructure, where failure carries physical and societal consequences, remains fragmented across disconnected technical, regulatory, and organizational vocabularies. Adopting a qualitative, theory-building design grounded in systematic document analysis, the study synthesizes peer-reviewed scholarship, standards documents, and regulatory texts published between 2016 and 2026 to identify convergences, tensions, and gaps in existing responsible artificial intelligence frameworks as applied to critical infrastructure. Thematic coding of forty primary sources reveals four recurring but insufficiently integrated concerns: governance and accountability, technical assurance, human-machine teaming, and regulatory compliance. Building on this synthesis, the paper proposes a layered conceptual framework that positions these four concerns as interdependent rather than parallel obligations, and maps the framework against the National Institute of Standards and Technology Artificial Intelligence Risk Management Framework, the European Union Artificial Intelligence Act, and the United States Department of Homeland Security roles and responsibilities guidance. The findings suggest that responsible artificial intelligence in critical infrastructure cannot be achieved through principle adoption alone; it requires structural coupling between engineering assurance practices and institutional accountability mechanisms. The paper concludes by outlining implications for infrastructure operators, standards bodies, and future empirical research.