When AI sees more than organizations can govern: Closing the accountability gap in radiology

AI in radiology is scaling diagnostic influence faster than hospitals are redesigning responsibility. Drawing on twenty-five semi-structured interviews with three senior radiologists and radiology leaders in the United States, Belgium, and Denmark, two executives from global medical-imaging and health-technology companies, and one executive from a major European health insurer, we examine why technically capable AI may remain confined to pilots and narrow workflows. The evidence shows that AI does not simply automate diagnostic work. It redirects professional attention, generates verification and documentation work, and creates new points of disagreement between algorithmic output and clinical judgment. We develop defensible reliance as the organizational condition in which professionals can use, question, or override AI in ways that are bounded, monitored, reconstructable, and institutionally justifiable. Four governance capabilities create this condition: task boundaries, lifecycle stewardship, accountability pathways, and legitimacy infrastructure. For hospital leaders, the central challenge is therefore not only whether AI performs accurately, but whether the decision system makes reliance on AI professionally and organizationally defensible.