MDR moves to Section B of the AI Act

A proposal is on the table to move the Medical Device Regulation (MDR) within the AI Act from Annex I Section A to Section B. The objective is clear: reducing the regulatory burden for manufacturers of medical devices with AI. That makes sense. For many parties, the current system feels like having to go through two compliance tracks simultaneously. First the MDR, and on top of that a substantial package from the AI Act. In this blog, I explain what this move means legally, why it is good news in practice, and what manufacturers should still be alert to.

Routes to high risk

The AI Act provides multiple routes through which an AI system can qualify as high risk. An important route is set out in Article 6(1). In short: AI is high risk if it is a safety component of a regulated product, or is itself a regulated product, and an independent third party is required for the conformity assessment. A medical device under the MDR can fall under this category.

This quickly leads to the heavy obligations for high-risk AI systems. Think of a risk and quality management system, technical documentation, logging, human oversight, requirements regarding accuracy, robustness and cybersecurity, and post-market monitoring. Some of these obligations strongly resemble what manufacturers must already arrange under the MDR. You then do not have one compliance route to follow; for medical AI, another compliance route is added on top.

What the move to Section B changes

By moving the MDR from Section A to Section B, medical devices with AI no longer automatically fall under the heavy obligations of Chapter III, Section 2 of the AI Act. The articles in this section relate to the requirements imposed on AI systems. At first glance, this seems strange; the AI Act is specifically about AI. Why would medical software with AI then be subject to fewer rules?

To make it slightly more complicated: there is also a proposal to amend the MDR. The answer lies in that proposal: the Commission will then have to take into account the requirements of Chapter III, Section 2 of the AI Act when drawing up certain instruments.

The clever workaround

What does this mean in practice? Take Article 14 of the AI Act on human oversight. This provision requires, among other things, that a human can understand, override and stop the output of an AI system. After the move to Section B, Article 14 no longer applies directly to medical devices with AI. However, the Commission can incorporate the essence of the article into a common specification. Think of a requirement that an AI-driven diagnostic device must always display a confidence score, so that the physician can assess how much weight to give to the outcome.

The same applies to Article 10 of the AI Act on data and data governance. Those requirements could be reflected in Annex I of the MDR through a delegated act, but tailored to the training data of medical AI. The commission translates the abstract AI Act requirements into concrete, sector-specific rules that fit the medical context.

In this way, the manufacturer no longer has to go through two separate compliance tracks, each with its own system, documentation and supervisory authorities. The AI Act requirements are not imposed directly, but integrated into MDR instruments that manufacturers already work with. The manufacturer therefore follows one route, the MDR route, and finds the relevant AI Act requirements there. A clever workaround that provides one integrated framework, sector-specific implementation of AI requirements, and less duplication of work.

Challenges for manufacturers

However, there are also caveats. First: "taking into account" is not a one-to-one adoption. The Commission is given discretionary room to determine how the AI Act requirements will be implemented in the MDR instruments. This means that manufacturers will have to wait and see what that translation will look like exactly.

Second: the Commission still has to draw up those instruments. Until then, there is uncertainty about which requirements precisely apply. Manufacturers already developing medical AI must therefore continue to monitor developments.

Third: for manufacturers active in multiple sectors, the move could actually lead to fragmentation. After all, the AI Act requirements are being implemented on a sector-specific basis in this case, which may mean that the same application must comply with different requirements in different contexts.

Conclusion: positive, with a caveat

The move to Section B is good news for manufacturers of medical devices with AI. The double compliance burden is eliminated. At the same time, manufacturers must remain alert to the concrete implementation by the Commission. The AI Act requirements are not disappearing; they are simply being repackaged.