Inter­play bet­ween MDR/IVDR and AIA in AI-assisted medi­cal devices (MDAI)

The docu­ment MDCG 2025–6 is a joint paper by the Euro­pean Arti­fi­ci­al Intel­li­gence Board (AIB)  and the Medi­cal Device Coor­di­na­ti­on Group (MDCG). It ser­ves as an FAQ docu­ment pri­ma­ri­ly for manu­fac­tu­r­ers of medi­cal devices with arti­fi­ci­al intel­li­gence (MDAI) and addres­ses the inter­ac­tions bet­ween the Medi­cal Devices Regu­la­ti­on (MDR), the In Vitro Dia­gno­stic Medi­cal Devices Regu­la­ti­on (IVDR) and the Arti­fi­ci­al Intel­li­gence Regu­la­ti­on (AI Act).

The aim is to pro­vi­de gui­dance to manu­fac­tu­r­ers, noti­fied bodies and com­pe­tent aut­ho­ri­ties on the simul­ta­neous appli­ca­ti­on of the­se regu­la­ti­ons, in par­ti­cu­lar with regard to high-risk AI sys­tems in medi­cal devices.

What it’s all about

Manu­fac­tu­r­ers of AI-based medi­cal devices will in future have to com­ply with the fol­lo­wing two com­ple­men­ta­ry sets of regu­la­ti­ons in particular:

• MDR/IVDR, which regu­la­te medi­cal safe­ty and cli­ni­cal effec­ti­ve­ness, and
• AI Act, which impo­ses addi­tio­nal requi­re­ments to ensu­re fun­da­men­tal rights, trans­pa­ren­cy, data qua­li­ty and risk manage­ment in AI systems.

When is an MDAI affected?

As soon as an AI sys­tem is a medi­cal device or part of one and must under­go con­for­mi­ty assess­ment by a Noti­fied Body in accordance with MDR/IVDR, it is also con­side­red a high-risk AI sys­tem under the AI Regulation.

MDAIs deve­lo­ped in-house by health­ca­re insti­tu­ti­ons are exempt from this, but must still meet cer­tain requi­re­ments of the AI Regulation.

What does this mean for manufacturers?

1. Qua­li­ty and risk manage­ment
   The AI Regu­la­ti­on pre­scri­bes a spe­ci­fic qua­li­ty manage­ment sys­tem (QMS) for high-risk AI sys­tems – this can be inte­gra­ted into the exis­ting MDR/IVDR QMS. Risk manage­ment must also cover AI-specific risks such as bias, data dis­tor­ti­on or unpre­dic­ta­ble sys­tem behaviour.
2. Data requi­re­ments
   AI trai­ning, vali­da­ti­on and test data must be of high qua­li­ty, traceable and repre­sen­ta­ti­ve of the tar­get popu­la­ti­on. Bias pre­ven­ti­on and moni­to­ring beco­me mandatory.
3. Trans­pa­ren­cy & over­sight
   Users must be able to reco­g­ni­se that they are inter­ac­ting with an AI sys­tem – explaina­bi­li­ty of how it works beco­mes a key cri­ter­ion. Human over­sight is man­da­to­ry: sys­tems must not be able to ‘over­ri­de’ themselves.
4. Tech­ni­cal docu­men­ta­ti­on
   Con­sis­tent tech­ni­cal docu­men­ta­ti­on for MDR/IVDR and AI Regu­la­ti­on is expec­ted. In addi­ti­on to cli­ni­cal evi­dence, AI-specific evi­dence (e.g. per­for­mance tests, risk assess­ments) must also be documented.
5. Post-market moni­to­ring & chan­ges
   Manu­fac­tu­r­ers must deve­lop inte­gra­ted moni­to­ring plans – also with regard to AI inter­ac­tions with other systems.

What does the inter­play bet­ween MDR/IVDR and AI Regu­la­ti­on mean for manufacturers?

• Two sets of rules – one goal: safe­ty, ethics and trust in AI-based medi­cal devices.
• Com­pli­ance will beco­me mul­ti­di­men­sio­nal in the future. Com­pa­nies must adapt to an inte­gra­ted, regu­la­ted life­cy­cle approach – from deve­lo­p­ment and data respon­si­bi­li­ty to ongo­ing mar­ket surveillance.
• Acting now means deve­lo­ping pro­ces­ses, docu­men­ta­ti­on and teams in a time­ly man­ner to ensu­re AI Regu­la­ti­on com­pli­ance. Only tho­se who address regu­la­to­ry issues ear­ly on will be able to bring MDAI pro­ducts to mar­ket safe­ly, eco­no­mic­al­ly and in com­pli­ance with the law.

We are hap­py to sup­port you in this endea­vour. Get in touch with us.