Tandem Health has extended its clinical AI from documentation into decision support — and done so inside a medical-device framework rather than as an unregulated feature. The detail worth noticing is not the capability itself, but the direction it marks. In 2026, clinical AI is splitting into two visible groups: tools that have committed to formal regulation and deepened their UK presence, and tools that have stepped back from the UK and EU rather than navigate the rules. Tandem is firmly in the first group.
What Tandem announced
Tandem says clinicians can now ask patient-specific clinical questions inside Ask Tandem during a consultation and receive sourced answers drawn from national guidelines and locally uploaded protocols. According to Tandem, this clinical decision support (CDS) product is CE marked under EU MDR as a Class IIa medical device and is available to selected customers across Europe and the UK. It sits in the same chat clinicians already use for documentation and draws on the context of the visit.
This builds on a deliberate climb: Tandem's AI scribe and coding assistant are also CE marked as Class IIa devices, externally assessed by a notified body. Decision support is the next rung.
What Class IIa actually means — and what it doesn't
Class IIa is a medical-device risk classification under EU MDR. Under Rule 11, software intended to provide information used for diagnostic or therapeutic decisions is generally Class IIa, escalating to IIb or III where the information's significance and the patient's condition raise the stakes. Class IIa requires independent assessment by a notified body against the product's intended use, with post-market monitoring attached.
Classification also turns on jurisdiction and intended purpose, not a ranking of safety. The same software can sit in different classes under different regimes — and the European Commission's December 2025 proposal would actually widen the range of lower-risk software that qualifies as Class I. A class number describes what a product is designed to do and where, not how rigorously it was built.
That is a genuine raising of the baseline. But the limit matters as much as the label: a regulatory class describes process and intended use, not a guarantee that every output is correct. NHS England's guidance on AI-enabled ambient scribing is explicit that AI-generated notes, summaries and letters can contain errors and must be checked by a professional before they enter the record. The same logic extends to decision support — certification raises the floor; it does not transfer clinical responsibility away from the clinician.
Three different jobs: scribe, coding, decision support
It helps to separate three things often lumped together as "AI in the consultation":
- A scribe documents the encounter.
- A coding assistant suggests diagnostic and procedure codes.
- Decision support offers information that shapes a clinical decision.
The risk profile rises across that list. A scribe that mistranscribes is usually caught at sign-off; decision support that subtly biases reasoning is harder to catch and carries more weight. That is why decision support attracts a higher regulatory class — and why "it uses AI" tells a buyer almost nothing. What the product does, and what it is certified for, is the question.
The 2026 divergence: who stayed and who stepped back
Tandem's move is most telling when set against the wider market. In the same period, several prominent clinical AI tools restricted UK access rather than operate under unsettled UK and EU rules:
| Tool | Type | Regulatory / UK status (2026) |
|---|---|---|
| Tandem CDS | In-consultation decision support | CE marked EU MDR Class IIa; available in UK/EU |
| OpenEvidence | Evidence search engine | Withdrew from UK and EU on 30 April 2026 |
| Heidi Evidence | Decision-support / evidence layer | Reportedly blocked UK NHS emails at enrolment (May 2026) |
| iatroX (AskIatroX) | Reference / knowledge layer | UKCA-marked Class I; remains available to UK clinicians |
The pattern is the signal. Regulatory clarity is becoming a deployment decision, not a back-office detail — and the tools that engaged with it (Tandem at Class IIa, iatroX at Class I) are the ones a UK clinician can still open.
Where iatroX sits
iatroX approaches the same underlying problem — reducing uncertainty and cognitive load around a clinical decision — from a different layer. Rather than living inside the documented consultation and shaping the record, iatroX is the clinician knowledge layer: AskIatroX surfaces fast, source-linked answers from validated UK guidance such as NICE, CKS, SIGN and the SmPC, alongside learning, exam preparation and clinical reasoning support, with the core clinical reference free to use.
AskIatroX is itself a UKCA-registered, MHRA-listed Class I medical device. It informs the clinician's own judgement rather than automating or directing the decision — which is precisely why it sits at Class I, below the Class IIa of in-consultation, patient-specific decision engines like Tandem's. Class I is self-declared rather than notified-body assessed, but self-declared is not ungoverned: iatroX's registration is backed by a complete clinical risk-management file and safety case, independently reviewed by an external advisor. That is a difference of scope, not a contest over labels: Tandem is building toward an EHR-adjacent operating system for the consultation, while iatroX is the trusted-answer and learning layer a clinician reaches for before, during and after care.
Tandem's announcement is a useful marker of where regulated, workflow-embedded clinical AI is heading. The wider lesson applies to every tool that crosses a clinician's desk: understand what it does, what it is certified for, whether it is available where you work, and what it still asks of your judgement.
Frequently asked questions
What is Class IIa under EU MDR? Class IIa is a medical-device risk classification under the EU Medical Device Regulation. Software providing information used for diagnostic or therapeutic decisions is generally placed here under Rule 11, requiring independent notified-body assessment against intended use plus post-market monitoring.
Does a Class IIa certification mean the AI is always correct? No. Certification is assurance about design, risk management, performance and intended use — not a guarantee every output is right. NHS guidance is clear that AI-generated clinical content can contain errors and must be checked before it is relied upon.
How is clinical decision support different from an AI scribe? A scribe documents the consultation; clinical decision support offers information that shapes a clinical decision. Decision support carries a higher risk profile, which is why it attracts a higher regulatory classification than documentation tooling.
Why have some clinical AI tools left the UK in 2026? Several US-origin tools, including OpenEvidence, cited regulatory uncertainty around AI systems — including the EU AI Act and an unsettled UK framework — as the reason for withdrawing from or restricting UK and EU access.
Is AskIatroX a regulated medical device? Yes. AskIatroX is a UKCA-registered, MHRA-listed Class I device. Class I is a lower risk class than Class IIa; the difference reflects that AskIatroX surfaces sourced guidance for the clinician to weigh, rather than shaping the record or issuing patient-specific instructions in-consultation. Class I is self-declared, but not ungoverned — iatroX maintains a full clinical risk-management file and safety case that has been independently reviewed.