PolicyIllustrative imageWhere the AI Act ends and the MDR begins for clinical software
Devices with machine learning components face two regulatory regimes at once, and the overlap is proving harder to manage than either framework alone.
A software device that uses machine learning to support a clinical decision is now subject to both the medical device regulations and the AI Act. The classification logic differs: the MDR class follows the intended purpose and the risk to the patient, while the AI Act assigns risk according to the use context and the potential for harm arising from the AI system itself. A triage tool can therefore be class IIa under one framework and high-risk under the other.
In practice most clinical AI falls into the high-risk category under the AI Act, which brings obligations for risk management, data governance, technical documentation, logging, human oversight, and accuracy. Several of these requirements rhyme with MDR Annex I but are not identical, and manufacturers must map the two sets item by item rather than assume that one satisfies the other. The data governance obligation, in particular, has no direct MDR equivalent.
The sequencing question matters for planning. Notified bodies assess MDR conformity, while market surveillance authorities enforce the AI Act, and the two do not always work to the same timetable. Manufacturers report that neither side yet has a settled view on how much of the AI Act evidence should sit inside the technical documentation, which produces inconsistent expectations across assessments and, in some cases, duplicate requests.
The pragmatic approach emerging among larger vendors is a single integrated risk file that carries both MDR and AI Act identifiers. It costs more to build initially, because the hazard analysis must address system-level failures such as distribution shift and biased training data, but it avoids maintaining two parallel descriptions of the same algorithm and makes post-market monitoring considerably easier to run across markets.
Disclaimer: this article is sample content for demonstration purposes. It is not regulatory advice, investment guidance or a clinical opinion. Always refer to the officially published legislation.
