AI Medical Scribe MDR Classification: Class IIa Is the Floor

On 27 May 2026, Sweden's Medical Products Agency (Läkemedelsverket) published the headline finding of its market surveillance inspection of an AI medical scribe. The Class I classification the manufacturer had been operating under was not compatible with EU MDR. Class IIa was the floor under Annex VIII Rule 11. The product had already been recertified at Class IIa by an independent notified body six days before the report, so the nonconformity was categorised as minor. The implication for the rest of the AI scribe category is anything but minor.

This raises a practical question:
What does Sweden's MPA decision on AI medical scribe MDR classification mean for every vendor and every healthcare organisation in Europe procuring one this week?

This article summarises what the MPA inspected and found, why an AI medical scribe lands in Class IIa under sub-rule 11(a) of MDCG 2019-11 rev. 1 rather than Class I under sub-rule 11(c), what Class IIa requires in practice, what the ruling means for the wider market and the five mistakes vendors keep making on classification. For the broader Rule 11 context, the companion piece on EU MDR Rule 11 software classification walks through the full framework.

Table of Content

What Sweden's MPA Just Decided

Sweden's MPA calls the underlying programme the AI Ambient Scribe Focus Initiative. Its objective is straightforward: assess whether AI scribes on the Swedish market are correctly qualified and classified under EU MDR. A preliminary inspection report of the first manufacturer in scope was published on 27 May 2026, with the agency's explicit permission.

Headline finding: one nonconformity (NC1) on classification, alongside four observations on documentation and labelling. The manufacturer had classified the AI medical scribe as Class I under Annex VIII Rule 11. According to the MPA, this is not compatible with MDR so the minimum applicable class is IIa.

The manufacturer had already recognised the issue and was preparing for Class IIa certification when the inspection began. Their class IIa certification was granted on 21 May 2026 by an independent notified body. That is why the nonconformity was categorised as minor because it had been remedied by the time the report landed. The substantive reasoning behind the decision is what every other AI medical scribe vendor in Europe needs to read this week.

-> In short: Sweden's MPA decided that an AI medical scribe is at minimum Class IIa under MDR. Class I is not compatible with the regulation and the reasoning applies across the category.

The Inspection: What Was Tested

The inspection scope covered the full picture an AI medical scribe presents to a regulator: product qualification and classification, intended purpose as stated and as observed in use, claims in marketing materials and the instructions for use, the relevant parts of the technical file, the product architecture and the functional mechanisms behind the inference. Plus a live demonstration and a functional test using simulated clinical conversations conducted by the MPA itself.

The AI Ambient Scribe Focus Initiative

The initiative is a proactive surveillance programme, not a complaint-driven investigation. The MPA explicitly designed it to assess the whole product category and the first vendor in scope was Tandem Health AB, which published the summary findings on 27 May.

The framework the MPA used is EU-wide. So other national competent authorities (BfArM in Germany, MHRA in the UK transition era, ANSM in France, AEMPS in Spain) operate under the same MDR and the same MDCG guidance. Which means a Swedish ruling on classification carries substantial weight for every EU member state.

One Nonconformity, Four Observations

The single nonconformity (NC1) was the classification call. Four observations (O1 to O4) covered intended purpose documentation, product distinction in user-facing materials, CE mark usage and declaration of conformity content. None of the four observations rose to nonconformity level.

That mix matters. The documentation and CE mark presentation gaps are common across the AI scribe category. Even a vendor who fixes the classification call still has documentation work to do and the MPA flagged these observations even after the classification was being corrected.

-> In short: The inspection scope was full. The findings split into one classification nonconformity (major in substance, minor in category because already being fixed) and four documentation observations.

Why Class I was not Enough Under Rule 11

The reasoning sits on two documents. The EU MDR Annex VIII Rule 11 sets the basic framework for medical device software classification. And the MDCG 2019-11 rev. 1, the European Commission's guidance approved by the Medical Device Coordination Group, fills in how the rule applies to specific software. MDCG 2019-11 rev. 1 splits Rule 11 into three sub-rules and the MPA's reasoning runs through them in order.

Sub-rules 11(a), 11(b) and 11(c)

Sub-rule 11(a) covers software intended to provide information used to take decisions with diagnostic or therapeutic purposes. The MDCG describes this as the mode of action characteristic of all medical device software (MDSW) and notes that sub-rule 11(a) is generally applicable to all MDSW except those with no medical purpose and the floor class is IIa.

Sub-rule 11(b) covers software intended to monitor physiological processes or parameters. The floor class is also IIa, escalating to IIb where vital parameters are at risk of immediate danger. Sub-rule 11(c) is the residual category: all other software and the only pathway to Class I. It applies to medical device software whose output does not influence clinical decisions and does not monitor physiological processes.

The "Continued Care Process" Doctrine

The MPA's reasoning hinges on a specific point. Even if no clinical decision is made during the consultation itself, the documentation the AI medical scribe produces is used as the basis for decisions in the continued care process and later visits draw on the note. Also other clinicians read it. Additionally care plans build on the recorded findings. So the software's output is used to take decisions with diagnostic or therapeutic purposes.

That reasoning is decisive. It moves the analysis from "does the AI make a decision" to "is the output used in decisions later". The latter is the test that puts the AI medical scribe under sub-rule 11(a) rather than sub-rule 11(c).

The Risk Management File Confirms the Class

The MPA's reasoning was reinforced by the manufacturer's own risk management file. The file identified, among other things, delayed diagnosis due to misinterpreted information as a credible risk scenario. By the manufacturer's own analysis, the software's output is capable of influencing clinical decisions.

That self-identified risk is the regulator's confirmation that the product falls under sub-rule 11(a). And it is hard to argue for Class I when your own risk file says misinterpretation can delay diagnosis. The point applies to any AI scribe whose risk register acknowledges similar harm pathways.

-> In short: Sub-rule 11(a) applies to information used in clinical decisions. The continued care process counts. The vendor's own risk file confirms the harm pathway exists.

What To Do:

Read MDCG 2019-11 rev. 1 § 4.4 on the Rule 11 sub-rules before settling on a class for any clinical-documentation AI.
Test your software against the "continued care process" question, not the "decision during the visit" question.
Cross-check your classification against your own risk management file. If the file lists clinical-decision harm scenarios, the class follows.

What an AI Medical Scribe Has to Show for Class IIa

Class IIa is a different operating model from Class I and the main difference is the notified body involvement. Beyond that, the QMS becomes externally audited, the technical file is reviewed in depth, the clinical evaluation has to be substantive and post-market surveillance becomes a structured ongoing programme. For a deeper view of how each class lands, the parent piece on medical device classes under MDR walks through all four.

Notified Body Path

Class IIa under MDR offers three conformity routes (Annex IX, XI Part A, XI Part B). For an AI medical scribe, Annex IX is the natural fit. It pairs a full ISO 13485 QMS audit with a technical documentation review on a sample. Software lives in design and lifecycle, not production batches, which is what Annex IX is built for.

The notified body market is tight. Audit slots for software-only Class IIa are running six to nine months out. Vendors who finalise their classification this week and start the notified body engagement now are looking at a certificate in 12 to 18 months. And vendors who wait will see that gap widen.

Technical Documentation Depth

MDR Annex II sets the technical file structure. For an AI medical scribe at Class IIa, the file has to cover the intended purpose precisely, the general safety and performance requirements (GSPR) check, a risk management file under ISO 14971, software lifecycle evidence under EN IEC 62304 (typically Class B), cybersecurity under EN IEC 81001-5-1 and a complete clinical evaluation report.

For AI specifically, the verification and validation pillar matters more than usual: models drift, training data shapes outputs. And the technical file has to document the dataset, the training and evaluation methodology, the performance metrics across patient populations and the post-market mechanism for monitoring drift.

Clinical Evaluation for Documentation Output

Clinical evaluation is a real piece of work for an AI scribe. The clinical claim is that the documentation output is accurate, reliable and safe for use in the continued care process and evidence has to support that claim. Comparative studies against clinician-only notes, error-rate measurements, qualitative review of clinical implications when the AI gets it wrong.

Equivalence routes are hard for novel AI medical scribes because there is no obvious predicate device with the same mode of action and the same clinical context. So most vendors should plan a small clinical investigation, designed around the AI medical scribe's specific intended purpose. The investigation does not have to be huge. But it has to be defensible.

-> In short: Class IIa means a notified body audit, a certified ISO 13485 QMS, a deeper Annex II file with IEC 62304 Class B software evidence and a substantive clinical evaluation.

What This Means for the Market This Week

The MPA decision lands hard for three groups: AI medical scribe vendors with products on the market or in development. EU healthcare organisations procuring or already using one. And the wider clinical AI category, where the same reasoning applies to anything generating documentation, summaries or decision-support content from clinical interactions.

For AI Scribe Vendors

If your AI medical scribe is currently classified or marketed as Class I, the position is now untenable. The MPA has stated explicitly that this classification is not compatible with MDR. Other competent authorities will follow the same reasoning because the framework is EU-wide.

The right move this week is to (1) re-classify against sub-rule 11(a), (2) audit your existing technical file against Annex II at Class IIa depth, (3) approach notified bodies to scope a Class IIa engagement and (4) communicate the recertification plan to existing customers before they hear it elsewhere. Waiting for a formal inspection letter is not a strategy.

For EU Healthcare Procurement

The relevant procurement question has shifted. It is no longer whether the vendor asserts the product is safe. Now it is about if the product has been independently assessed as safe and performant, which is the requirement that comes with a Class IIa CE mark. So a Class I AI medical scribe is, after this ruling, a procurement risk.

For organisations already deploying one, the practical step is to ask the vendor for their Class IIa certification timeline. And for new procurements, the right requirement is a Class IIa CE mark or a credible recertification path.

For Other Clinical AI Categories

The reasoning generalises: Any clinical AI that produces information used in the continued care process is exposed to the same sub-rule 11(a) analysis. Ambient summarisation tools, AI coding assistants whose suggestions feed back into the patient record, AI-generated discharge summaries, AI handover notes. All of these share the structural feature the MPA used to classify the AI medical scribe at IIa.

So the Tandem inspection is the precedent the rest of the clinical AI documentation category will be measured against. The category needs to plan for IIa as the default position, with IIb or III only on case-specific risk escalation.

-> In short: Vendors should re-classify and approach notified bodies this week. Procurement should require Class IIa. The reasoning applies to the wider clinical-documentation AI category.

What To Do:

Vendors: re-classify against sub-rule 11(a), audit the technical file against Annex II at IIa depth, approach notified bodies this week.
Procurement: require a Class IIa CE mark or a documented recertification timeline. Treat Class I as a contract risk.
Other clinical-AI categories: assume the same sub-rule 11(a) reasoning applies to your product and validate the classification call against the continued care process test.

Common Mistakes on AI Medical Scribe MDR Classification

From the AI scribe space and the wider clinical AI category, five classification mistakes come back consistently. The Tandem inspection report surfaced four of them across its observations. And the fifth shows up at notified body engagements.

1. Evaluating Against the "During the Visit" Test

Vendors argue that no clinical decision is made during the consultation, so the AI scribe falls under sub-rule 11(c). The MPA's reasoning rejects that frame and the test is whether the output is used in the continued care process. For an AI medical scribe writing into the record, it is.

2. Claiming the Clinician Validates Everything

"The clinician reviews and signs the note" is a common defence and does not move the classification. The clinician's review is a human-in-the-loop control on the software's output. The software is still providing information used in decisions and controls do not exempt the device from classification.

3. Mixing the Intended Purpose and the Marketing Message

Marketing copy positions the product as a productivity tool, while the intended purpose statement quietly describes a clinical function. And these two have to align since regulators reads both. Tandem's inspection flagged this as observation O2: product distinction in user-facing materials.

4. Self-Applied CE Mark Without the Right Route

Class I lets the manufacturer self-declare. Once the class moves to IIa, the CE mark needs a notified body number alongside it. Vendors who continue to display the CE mark in Class IIa form before the notified body has signed off are misrepresenting their certification.

5. Ignoring MDCG 2019-11 rev. 1 in the Classification Record

The classification record in the technical file should walk through MDCG 2019-11 rev. 1 step by step. Qualification first, then the sub-rule analysis under Rule 11. Records that cite Rule 11 without engaging with the sub-rules invite a regulator to do the analysis themselves and reach a different conclusion.

-> In short: Wrong test frame, human-in-the-loop defence, marketing-IFU mismatch, premature CE mark, missing MDCG analysis. Five paths to a classification nonconformity.

FAQs About AI Medical Scribe MDR Classification

Q1: Can an AI medical scribe be Class I under MDR?

A1: No. Sweden's Medical Products Agency has stated explicitly that Class I is not compatible with EU MDR for an AI medical scribe whose output is used in the continued care process. The minimum class under Annex VIII Rule 11, sub-rule 11(a), is IIa and higher classes apply where the risk profile warrants.

Q2: What if the clinician edits and signs the note?

A2: Clinician review is a control on the software's output. But it does not exempt the software from classification. The MDSW provides information used in clinical decisions whether or not a clinician validates the output. So clinician sign-off does not move the AI medical scribe out of sub-rule 11(a).

Q3: Does the decision apply outside Sweden?

A3: The MPA's reasoning is anchored in EU MDR and MDCG 2019-11 rev. 1, both of which apply across the EU. Other national competent authorities (BfArM, ANSM, AEMPS) operate under the same framework. So while only Sweden has issued this specific ruling so far, the same analysis is available to every EU regulator and is the foundation any sponsor should plan against.

Q4: How long does Class IIa certification take?

A4: Plan for 12 to 18 months from project start to a Class IIa certificate. The biggest variable is the notified body audit slot, which is currently running six to nine months out for software-heavy IIa portfolios. Vendors who have the technical file substantially in place can compress that. Vendors starting cold should plan against the longer end.

Closing Thoughts

The AI medical scribe MDR classification question has been a source of category-wide uncertainty for two years and Sweden's MPA closed it. Class IIa is the floor under sub-rule 11(a). The reasoning applies to every AI scribe in the EU and to the wider clinical-documentation AI category. So the market shift this triggers will play out over the coming months as other competent authorities publish their own findings and as procurement teams revise their requirements.

For vendors who need to move from Class I to Class IIa, or to verify that a current Class IIa positioning is defensible, a structured MDR consulting engagement is the fastest way to align the classification, the technical file and the notified body path before the next inspection letter arrives.