Article 4 and Article 72 in radiological practice: competence and surveillance beyond compliance

Two provisions of the EU Artificial Intelligence Act (Regulation (EU) 2024/1689) now reach into radiological practice. Article 4, applicable since February 2025 and enforced from August 2026, requires a sufficient level of AI literacy in providers and deployers. Article 72 will require a post-market monitoring system over the lifecycle of high-risk AI systems; for medical AI they are set to take effect on 2 August 2028 under the amended timeline. Current compliance framework: The European radiological community has responded with a coherent body of work: the ESR Essentials series, statements from the ESR AI Working Group and a multi-society coalition, the ESR post-market-surveillance consensus, and a recent narrative review of implementation errors.

This Opinion Paper is not a regulatory commentary. Two operational shifts are now due. First, Article 4 should be read as a floor - the legal minimum - over which an audit-capable competence framework is built: radiologists trained not to recite failure modes but to detect them in models deployed on their own population. Article 72 should be operationalized not as a documentary monitoring plan filed with notified bodies but as the architecture of a clinical surveillance culture analogous to pharmacovigilance, with continuous local drift detection, subgroup audit, public reporting, and shutdown criteria.

Neither shift is technically novel; their value lies in the explicit regulatory anchoring. The distinction between compliance and culture is the operational frontier of clinical AI in European radiology.