Governing the Invisible: Cognitive Externalities of AI and the Limits of Current Accountability Frameworks

Current AI governance frameworks are built to address harms that are visible, measurable, and immediate—bias in automated systems, privacy violations, and misinformation at scale. This paper argues that these frameworks are structurally blind to a category of harm that is equally significant but far harder to govern: the gradual erosion of human cognitive capability through habitual AI dependence. Drawing on emerging research in cognitive offloading, desirable difficulty theory, and the developmental psychology of creative cognition, this paper proposes the concept of cognitive externalities—indirect, diffuse, and long-term costs that current accountability mechanisms are not designed to capture. It further argues that this gap is not incidental but structural: our governance vocabulary was developed for harms with clear causal chains and identifiable victims. Cognitive externalities have neither. The paper concludes by proposing a preliminary framework for how policymakers, educators, and technologists might begin to address this governance gap.