As clinical AI matures, more tools now advertise a medical-device classification — Class I, Class IIa, occasionally IIb or III. It is becoming a marketing line as much as a regulatory fact, and it is widely misread. A higher number is often taken to mean a better, safer or more sophisticated product. That is not what the classification system says. A device's class encodes its intended purpose and the jurisdiction it was certified in — not its sophistication, and not a league table of safety. This guide explains what the badge actually means, so you can read vendor claims accurately rather than taking the number at face value.
What the classes mean
Under the EU Medical Device Regulation, software is classified by Rule 11 according to what it is designed to do and the consequences if it gets something wrong. In broad terms:
| Class | Typical software | Conformity route |
|---|---|---|
| Class I | Stores, communicates or provides reference information for a clinician to interpret | Self-declared |
| Class IIa | Provides information used to inform a diagnostic or therapeutic decision | Notified-body assessed |
| Class IIb | Drives clinical management in more serious situations | Notified-body assessed |
| Class III | Informs decisions in life-threatening conditions | Notified-body assessed |
The UK runs on similar principles under its own framework, and its 2026 reforms move it closer to this international model.
The principle most people miss: class follows the job, not the quality
Here is the key to reading these badges. The class follows the intended purpose. A tool designed to retrieve and present trusted information for a clinician to weigh is Class I. A tool designed to interpret, calculate or shape a patient-specific decision — or to write to the medical record — is Class IIa or higher. The number tells you which of those a tool is; it does not tell you how well the tool does its job. An excellent Class I tool and a mediocre Class IIa tool are both entirely possible. The class does not rank them against each other.
A worked spectrum
It is easiest to see by mapping real categories — not as a ranking, but as different jobs that correctly attract different classes:
- A record-shaping ambient scribe or coding assistant writes to or codes the medical record, so it sits at Class IIa. Tandem, for example, has taken its scribe and coding products to Class IIa — the appropriate class for software that shapes the record, and a genuinely demanding standard to meet for that job.
- A citation-first reference and knowledge tool surfaces sourced answers for the clinician to interpret and verify, so it sits at Class I. AskIatroX is an example: it informs the clinician's own decision rather than making it.
- A generative diagnostic decision-support tool that proposes a diagnosis or management plan sits higher, typically at Class IIb.
Same broad field, three different jobs, three different classes. None is "better" by virtue of its number; each class is correct for what the tool is built to do.
"Self-declared" is about the route, not the rigour
Because Class I is self-declared — no notified body signs it off — it is sometimes framed as "less governed." That conflates the conformity route with the quality system behind it. A serious Class I clinical product still runs a clinical risk-management process, with a hazard log and a clinical safety case, and a responsible manufacturer has that documentation independently reviewed. iatroX's Class I registration, for example, is backed by a complete clinical risk-management file and safety case that has been independently reviewed by an external advisor. "Self-declared" describes who signs the certificate, not whether the work was done.
The same software can be classified differently in different places
Classification is also jurisdictional. Under EU MDR Rule 11, software that informs diagnosis or treatment generally lands at Class IIa, and the European Commission's December 2025 proposal would actually widen the range of lower-risk software that can be Class I. The UK applies similar principles and is aligning more closely with that international framework. So a class number is partly a statement about where a tool was certified — not a universal score that travels unchanged across borders.
What to ask instead of reading the number
A class is a starting point for questions, not a verdict. When you see one, ask:
- What is the tool designed to do — reference, document, or decide? That is what determines the appropriate class.
- What does its intended-purpose statement say? That is the basis on which the class was assigned.
- What clinical safety standard sits behind it — is there a clinical risk-management file and safety case, and has it been reviewed?
- Where do its answers come from? For a reference tool, source provenance — named, current, trustworthy sources — is where safety actually lives.
The badge tells you the category. These questions tell you the quality.
Where iatroX sits
AskIatroX is a citation-first clinical reference and knowledge layer: it surfaces source-linked answers from validated UK guidance — NICE, CKS, SIGN and the SmPC — for the clinician to interpret and verify. That intended purpose is exactly why it is a UKCA-registered Class I device: it is built to inform the clinician's decision, not to make it. Its Class I status is backed by a full, independently reviewed clinical risk-management file, and the core clinical reference is free to use. The class is not a ceiling on rigour — it is the correct label for a tool designed to inform rather than decide.
Higher number does not mean better tool. It means a different job. Read the class for what it is — a statement of intended purpose and jurisdiction — and judge a clinical AI tool on what actually makes it trustworthy: what it is for, how it is governed, and where its knowledge comes from.
Frequently asked questions
What does Class I or Class IIa mean for a clinical AI tool? It describes the tool's intended purpose and risk, not its quality. Class I generally covers software that stores, communicates or provides reference information for a clinician to interpret; Class IIa covers software that provides information used to inform a diagnostic or therapeutic decision, and requires assessment by a notified body.
Is a Class IIa tool safer or better than a Class I tool? Not inherently. The class reflects what a tool is designed to do, not how well it does it. A Class IIa scribe and a Class I reference tool are doing different jobs, and each class is the correct one for its job — the number is not a ranking of quality between them.
Does "self-declared" Class I mean the tool is unregulated? No. Self-declaration is the conformity route — the manufacturer attests to meeting the requirements rather than a notified body signing off. A serious Class I product still maintains a clinical risk-management file and safety case, ideally independently reviewed. "Self-declared" is about who signs the certificate, not whether the safety work was done.
Why is a clinical reference tool classified as Class I? Because its intended purpose is to surface information for a clinician to interpret and act on, rather than to make or direct a patient-specific decision. Under Rule 11, software that provides reference information without interpreting it for the decision generally sits at Class I in both the EU and UK frameworks.
Can the same software have a different class in different countries? Yes. Classification depends on jurisdiction as well as intended purpose. EU MDR Rule 11 and the UK framework apply similar principles, but the specific class — and proposed reforms such as the European Commission's December 2025 changes — can differ, so a class is partly a statement about where a tool was certified.