From Covert Use to Structured Oversight: A Framework for Institutional Governance of Artificial Intelligence in Graduate Medical Education

Problem Statement: Artificial intelligence (AI) tools are routinely used by residents and fellows across graduate medical education. The Accreditation Council for Graduate Medical Education has no AI-specific policy nor requirement for disclosure norms, leaving programs and attending physicians without a standard framework for supervising AI-assisted trainee work. The result is a training environment in which AI use is simultaneously widespread and under-governed by operating without the policies and supervisory frameworks needed to make that use educationally productive. Background: AI tools are now embedded across health system workflows, and their use has expanded beyond clinical documentation into education, research, and scholarly work. At MaineHealth, a gap became visible: as AI entered our clinical environment through institutional platforms and trainee-driven adoption alike, our program found itself without a shared framework for discussing, supervising, or teaching around its use. Application: Drawing on a structured debate at the Roux Institute at Northeastern University and the emerging graduate medical education literature on AI supervision, we propose a practical governance framework for AI use in residency and fellowship training. The framework synthesizes current evidence and existing supervisory structures to offer guidance across 4 domains: task-level expectations for trainees, real-time supervision strategies for attending physicians, integration of AI literacy into existing didactic structures, and program-level governance principles. AI literacy is a core professional competency, not optional enrichment. We must address not whether to engage with AI, but how, and in ways that preserve learning and build habits appropriate for AI-enabled practice.