Legal Frameworks for Governing Generalist and Agentic Artificial Intelligence in Medical Imaging

Artificial intelligence (AI) in medical imaging is moving from narrow algorithms toward generalist systems that perform multiple tasks and agentic workflows that plan, invoke tools or specialist models, and initiate downstream actions. This transition exposes a mismatch between what current AI and medical-device regulation commonly evaluates: a defined product with an intended use and the variable human-AI workflows that increasingly produce clinical consequences. We propose a five-layer accountability framework spanning data, model, orchestration, clinical use, and institutional lifecycle to demonstrate how familiar legal and governance principles can be translated into auditable controls. Clinically consequential uses will initially require meaningful clinician oversight, but this should not make the clinician the only, or even the default, bearer of responsibility for all upstream failures. Generalist and agentic AI systems may eventually introduce a varying amount of autonomy in healthcare systems, requiring us to rethink existing frameworks of clinical AI governance, with institutions, developers, and insurers requiring new role-specific obligations and coverage models calibrated to the degree of autonomy across the full decision pathway.