AI Regulation Tracker / Signal
IMDRF Closes Public Consultation on a Draft Framework for AI Life Cycle Management in Medical Devices
On July 11, 2026, the International Medical Device Regulators Forum closed the public comment period on draft document N93, a proposed technical framework for managing AI across the medical device life cycle. The draft is out for comment, not adopted. It is non-binding, and no regulator has implemented it.
On July 11, 2026, the International Medical Device Regulators Forum closed the public comment period on a draft it had been circulating since April. The document is proposed document N93, a "Technical Framework for Artificial Intelligence Life Cycle Management." It was published through IMDRF's Artificial Intelligence and Machine Learning-enabled Working Group, whose secretariat is hosted by the US Food and Drug Administration's Center for Devices and Radiological Health. The consultation ran on the forum's Citizen Space platform, opened April 11, 2026, and closed on July 11, 2026.
What the draft framework sets out to do
N93 is meant to give regulators and manufacturers a common way to think about AI across the whole life of a device, not just at the moment it is approved. That is the part worth slowing down on. A lot of the difficulty with AI-enabled medical devices is that the software does not stand still. It gets trained, retrained, updated, and monitored, and the questions that matter at each stage are different. A framework that covers the total product life cycle tries to give one structured view of all of that.
The draft groups its considerations into a few areas that come up again and again in this field. Data quality is one, meaning where the training and evaluation data came from and whether it is fit for the job. Model transparency is another, meaning how well the behavior of the model can be understood and explained. Performance evaluation is a third, covering how a device is tested before and after it reaches the market. And running underneath all of it is the point about multidisciplinary expertise, the idea that overseeing these systems takes clinicians, data scientists, engineers, and quality people working together rather than any one function on its own.
The stated aim is alignment. IMDRF is a forum where medical device regulators from different countries compare notes and try to converge on common approaches. A harmonized framework for AI-enabled devices would let manufacturers work against one shared set of considerations rather than reconciling several national interpretations from scratch. That is the destination the draft is pointed at. It is not the place the draft has arrived.
Where this sits in the process
N93 does not stand alone. It builds on an earlier IMDRF document, N88, the good machine learning practice guiding principles, which was finalized in 2025. Those principles set out high-level expectations for developing machine learning-enabled devices. N93 is the next layer, an attempt to turn that thinking into a fuller life cycle framework.
The important word is draft. This document was published for public consultation, which is the stage where a regulator or a forum puts a proposal in front of industry and the public and asks for reactions. That window has now closed. What happens next is that the working group takes the comments it received and revises the text, working toward a version that could eventually become a final harmonized framework. There is real distance between a closed consultation and a finished document, and more distance still between a finished IMDRF document and anything a national regulator formally adopts.
What this is, and what it is not
I want to be careful with the language here, because it is easy to see a group of regulators and a serious-sounding title and assume something has been decided.
Nothing has been decided. N93 is a proposed document that was out for comment. The consultation closing does not turn it into a standard, a rule, or a requirement. IMDRF itself does not regulate any market. It is a voluntary forum, and its outputs are guidance that individual regulators may or may not choose to reflect in their own systems, on their own timelines. No regulator has implemented this framework, because there is no final framework to implement yet.
What the close of the consultation does tell you is direction. When the medical device regulators who sit in this forum spend time drafting a total-lifecycle framework for AI, and organize it around data quality, model transparency, performance evaluation, and multidisciplinary oversight, that is a clear read on the themes they think matter. It is a weathervane for where oversight of AI-enabled devices is heading. It is not a wind anyone is legally required to follow today.
What this means for US professionals
The US angle is worth naming plainly. FDA CDRH hosts the secretariat for the working group that produced this draft, so the agency is close to the center of the drafting, not a bystander reading it later. And the track record matters. Earlier IMDRF work in this area, including the good machine learning practice principles, has historically fed into how FDA talks about and frames its own expectations for AI and machine learning software as a medical device. That is why a closed consultation on N93 is a genuine signal for a US audience rather than distant international housekeeping.
If you work in or around regulatory affairs, quality, or AI governance for medical devices, the value now is preparation, not compliance. Reading the draft tells you the vocabulary and the structure the regulators are gravitating toward for total-lifecycle oversight. Mapping your own AI governance against those four areas, data quality, transparency, performance evaluation, and multidisciplinary expertise, is a reasonable way to get ahead of where finalized expectations may land. But do it as forward planning, not because any rule obliges you to.
For US CPAs, finance, and broader compliance leaders watching AI regulation across sectors, the lesson generalizes. This is another case of a total-product-lifecycle view winning out over a one-time approval view. Regulators keep landing on the same idea, that AI systems have to be governed continuously, with clear ownership, documented data provenance, and evidence you can produce on demand. That posture travels well beyond medical devices, and it is the direction to plan against.
What to do now
Read the draft as a draft, not as a rule. The correct takeaway is that a consultation on a proposed AI life cycle framework closed on July 11, 2026, not that a global standard has been set. If you touch AI-enabled devices, review N93 and check your governance against its four core areas while there is still time to influence the final text and to prepare. Keep a named owner accountable for the AI life cycle in your programs, from data sourcing through post-market monitoring. Write down your answers on data provenance, model transparency, and performance evaluation before a regulator asks for them. And do not change a control or a submission on the strength of this draft alone. Base any change on the standards that actually apply to you and on advice from qualified regulatory professionals.
Questions professionals are asking
Did IMDRF issue a new rule for AI in medical devices?
No. IMDRF published a draft document, N93, for public consultation, and that comment period closed on July 11, 2026. The draft is non-binding and out for revision. IMDRF is a voluntary forum that does not regulate any market. The document creates no rule, standard, or legal duty, and no regulator has implemented it.
What does the draft framework cover?
It sets out considerations for managing AI across the full medical device life cycle, organized around data quality, model transparency, performance evaluation, and the multidisciplinary expertise needed to oversee these systems. It builds on IMDRF's finalized N88 good machine learning practice principles and aims to help align international oversight of AI-enabled devices.
Is the framework final or adopted?
No. The public consultation on the draft closed on July 11, 2026, and the document now goes back to the working group for revision toward a possible final version. It is not finalized, not adopted, and not binding, and no regulator has put it into force.
Why does this matter for US professionals?
The US FDA's Center for Devices and Radiological Health hosts the secretariat for the working group that produced the draft, and prior IMDRF outputs, such as the good machine learning practice principles, have historically been reflected in FDA guidance. So the draft signals where FDA's total-lifecycle expectations for AI and machine learning software as a medical device may be heading.
Should we change our AI governance because of this draft?
Only as forward planning. The draft can help you benchmark your controls against the four areas it emphasizes, but any change to your controls, documentation, or submissions should rest on the standards that actually apply to you and on advice from qualified regulatory professionals, not on a draft that is still under revision.
RELATED BRIEFINGS
Browse the full AI Regulation News tracker
Informational analysis for working professionals, not legal, regulatory, or medical device advice. Confirm how any standard or requirement applies to your situation with qualified professionals in the relevant jurisdiction.