With TORTUS reaching UKCA Class IIa and Tandem reaching EU MDR Class IIa within weeks of each other, the direction of travel looks clear: as AI scribes move beyond transcription into summarisation, coding and decision support, Class IIa is starting to look like the expected bar rather than a differentiator. The important caveats are that there is no settled benchmark yet for AI documentation tools, and the ceiling, where a scribe becomes a decision-support device, is still being drawn. Here is the trend and where it may be heading. The prediction section at the end is clearly framed as opinion.
Key takeaways
- As scribes add summarisation, coding and downstream actions, their risk classification rises.
- Two leading tools reaching Class IIa signals a shift from differentiator to expectation.
- The MHRA's AI Airlock is actively working on where documentation crosses into decision support.
- NHS England's framework already sets at least Class I registration as a floor for deployment.
- The boundary between a scribe and a decision-support device is the key unresolved question.
What is driving tools up the risk ladder?
Function drives class. A tool that only transcribes speech is generally not a medical device. Once it summarises beyond simple transcription, it typically becomes a Class I device. When it goes further, suggesting clinical codes, drafting referrals, or feeding into downstream ordering, it starts to influence clinical decisions, which pushes it toward Class IIa. The market has moved decisively in that direction: buyers want scribes that code and draft and integrate, not just transcribe. Each of those capabilities adds clinical risk, and risk is what determines classification. So the feature race and the regulatory ladder are pointing the same way.
The AI Airlock problem
The regulatory challenge is that the boundary is genuinely unclear. The MHRA's AI Airlock programme, with TORTUS selected into Phase 2, is specifically examining where an AI documentation tool crosses from documentation support into diagnostic or decision-support functionality. Under the UK Medical Device Regulations, a change in intended use can trigger reclassification from Class I to Class IIa, but there is no established benchmark for AI-based clinical documentation systems. That uncertainty is precisely what the Airlock work is trying to resolve, by exploring practical approaches to verification, validation and post-market surveillance for higher-threshold, LLM-enabled functionality.
What floor does NHS England's framework set?
The NHS is not waiting for the boundary to be perfectly drawn. NHS England's 2025 guidance on ambient scribing, developed with the MHRA, confirmed that scribes with certain functionalities fall under the UK Medical Devices Regulations 2002 and must be MHRA registered. The NHS England AVT Supplier Registry then set a practical floor for deployment: at least MHRA Class I registration, a completed DTAC assessment, and a commitment to post-market surveillance. So there is already a minimum bar, and the direction of expectation is upward as higher-risk functionality becomes normal.
What happens to unregulated or Class I scribes if IIa becomes the expectation?
If buyers come to expect Class IIa for tools that code and draft and influence decisions, several things follow. Unregistered tools become hard to procure and carry real liability risk for the clinicians using them, since using an unregistered medical device shifts risk onto the deployer. Class I tools that genuinely stay within summarisation may remain appropriate for that narrower purpose, but tools that quietly add higher-risk features while remaining self-declared will face pressure to certify. The likely effect is a sorting of the market: tools that match their certification to their actual functionality, and tools that do not.
Where is the ceiling? When does a scribe become a decision-support device?
This is the open question. Class IIa already covers decision-support functionality, and TORTUS has been certified as a clinical decision-support device. But there is a further boundary above that, where software influences decisions in ways that could cause serious harm if wrong, which pushes toward higher classes still. The more a scribe does than document, suggesting management, flagging findings, prompting actions, the closer it moves to being a decision-support device in the fuller sense, with the heavier assurance that implies. Drawing that line clearly is the substance of the current regulatory work.
Prediction (opinion)
The following is opinion rather than established fact. Over the next twelve months, I would expect more scribes to pursue and achieve Class IIa, buyers to start asking for it by default for tools that code and draft, and the AI Airlock work to produce clearer expectations for validating higher-risk documentation functionality. I would not expect Class I to disappear, since genuinely narrow summarisation tools have a place, but I would expect the gap between a tool's certification and its real functionality to become a procurement red flag. In short, device-grade is likely to shift from a selling point to a baseline for anything beyond pure transcription. For where each tool sits now, see our AI scribe regulatory status tracker, and for the classes, our guide to device classification.
Frequently asked questions
Is Class IIa becoming the standard for AI scribes? It is trending that way. As scribes add summarisation, coding and decision-support features, their risk rises, and two leading tools reaching Class IIa suggests it is becoming an expectation rather than a differentiator. There is not yet a formal benchmark.
Why do coding and drafting features raise a scribe's classification? Because they influence clinical decisions and records, which adds risk. Transcription alone is usually not a device; summarisation is typically Class I; decision-influencing functions push toward Class IIa.
What is the MHRA AI Airlock doing about this? It is examining where AI documentation tools cross into decision support, and how to validate higher-risk, LLM-enabled functionality. TORTUS is in Phase 2, focused on intended use and validation.
Can a Class I scribe still be used in the NHS? Yes, within its intended purpose and NHS England's framework, which sets at least Class I registration as a floor. But tools that add higher-risk features while remaining self-declared will face pressure to certify at Class IIa.
Will unregistered scribes become a liability? Increasingly, yes. Using an unregistered medical device shifts risk onto the deployer and clinician. As expectations rise, unregistered tools become harder to procure and riskier to use.