Executive overview
As artificial intelligence becomes more integrated into UK clinical practice, understanding the regulatory landscape is no longer optional—it's essential for patient safety and professional accountability. The MHRA’s AI Airlock is now live and moving to a second phase, providing a supervised sandbox for innovative AI as a Medical Device (AIaMD). In parallel, new Post-Market Surveillance (PMS) regulations, effective from 16 June 2025, place stricter duties on all device manufacturers, including software. For NHS buyers, the DTAC and NICE ESF frameworks remain the baseline for procurement and evidence standards.
This article provides a practical guide to this new reality through four distinct case studies, showing how different AI-driven platforms are engaging with regulation:
- AutoMedica SmartGuideline: A guideline retrieval tool being tested in the MHRA AI Airlock.
- Prof. Valmed: A generative AI clinical decision support system with a high-risk EU MDR Class IIb CE mark.
- iatroX: A UKCA Class I, MHRA-registered clinical reference and Q&A app.
- TORTUS: An ambient scribe with MHRA Class I registration, piloted in NHS settings.
Primer: how UK regulation actually works
- Registration: A medical device must be registered with the MHRA before it can be placed on the Great Britain market. Non-UK manufacturers must appoint a UK Responsible Person.
- Classification: Devices are classified by risk. Class I devices (e.g., many informational tools) can typically be self-declared by the manufacturer, while higher-risk classes (IIa, IIb, III) require a formal conformity assessment by a UK Approved Body.
- NHS Assurance: The DTAC (Digital Technology Assessment Criteria) is the baseline for procurement into the NHS, covering safety, privacy, and security. The NICE ESF (Evidence Standards Framework) sets the standards for the clinical and economic evidence a tool needs to provide.
- New GB PMS Rules (from 16 June 2025): These new regulations mandate tighter incident trending, require Periodic Safety Update Reports (PSURs), and enforce stricter vigilance expectations for all devices on the market.
Case studies (what the claims actually mean)
AutoMedica SmartGuideline — sandboxing AI guideline search
- Status: AutoMedica’s SmartGuideline was selected for the first pilot cohort of the MHRA AI Airlock. The company is exploring a "TAG-RAG" (Tagged Retrieval-Augmented Generation) approach to ground its answers in trusted content from NICE, CKS, and the BNF.
- Implication for clinicians: This tool is undergoing early, supervised real-world testing with direct regulator oversight. The key benefit of the Airlock is that learnings will be published, providing transparent insights into the safety and efficacy of this type of AI before any wide-scale rollout.
Prof. Valmed — EU MDR Class IIb CE mark for LLM-enabled CDS
- Status: In a significant milestone, Prof. Valmed has announced that its generative AI-powered clinical decision support system has achieved a Class IIb CE mark under the EU Medical Device Regulation (MDR). This is one of the first instances of a complex, LLM-based tool meeting such a high regulatory bar in Europe.
- Implication for UK readers: While CE-marked devices can support a route to the Great Britain market, they must still be registered with the MHRA and are subject to UK rules, including the new PMS regulations. Any UK organisation looking to adopt such a tool would still need to conduct its own local governance checks, including a DTAC assessment.
iatroX — UKCA Class I, MHRA-registered clinical reference/Q&A
- Status: iatroX is a UKCA Class I, MHRA-registered medical device. Its intended purpose is to act as a clinical reference and educational tool, providing citation-first answers to questions and structured brainstorming aids.
- Implication: This classification makes it suitable for education and reference workflows where clinicians can verify the cited sources and last-updated dates. Its use in practice should always be in line with local Trust policy for informational tools.
TORTUS — ambient voice technology (AI scribe)
- Status: TORTUS is an ambient scribe that has been piloted in NHS settings, including at Great Ormond Street Hospital. The company reports it has an MHRA Class I registration, with work underway to upgrade to Class IIa.
- Policy backdrop: The use of any AI scribe in the NHS must adhere to the April 2025 NHS England guidance, which sets clear expectations for clinical safety, data protection (DPIA), and mandatory human verification of all outputs. The BMJ has also issued a strong warning against the use of any unregistered scribe tools.
The wider landscape
The UK is building a joined-up ecosystem for AI regulation. The AI Airlock provides a path for novel devices, while the AI & Digital Regulations Service, a multi-agency group led by NICE, offers a "one-stop shop" to help developers and NHS adopters navigate the complex pathway from idea to implementation.
How to compare regulated AI tools (buyer’s checklist)
- Intended purpose & class: What is the tool designed to do? Is it Class I, IIa, or IIb? Check its MHRA registration status on the Public Access Registration Database (PARD) where available.
- Evidence posture: What NICE ESF tier does it meet? Is there a published external evaluation? Does it provide clear provenance (citations, last-updated dates)?
- PMS readiness: Does the vendor have a clear plan for meeting the new Post-Market Surveillance rules after 16 June 2025, including incident capture and trend analysis?
- NHS readiness: Has the vendor passed a DTAC assessment? Is a DPIA and a clinical safety case (DCB0129/0160) available? Are data flows and audit logs clear?
Practical micro-workflows
- Guideline lookup with provenance: A doctor on an outpatient ward queries a citation-first tool like iatroX to get a quick, referenced answer. They follow the links to the original NICE/CKS guideline to verify, then paste the citations into their notes.
- Clinic documentation: A GP in primary care uses an approved ambient scribe like TORTUS. They obtain patient consent, allow the tool to draft the note, then personally verify every line before saving it to the record. They report a minor transcription error via the vendor's PMS feedback channel.
- Introducing a higher-risk CDS: A hospital Trust considering a CE-marked, Class IIb tool like Prof. Valmed first confirms its GB registration with the MHRA, completes a full DTAC assessment, and establishes a local Standard Operating Procedure (SOP) before any live clinical use.
Risks & Mitigations
- Regulatory drift & scope creep: Ensure the tool's intended purpose in its official labelling matches how you are using it. If a vendor adds new features, this may trigger a re-classification, so re-assess accordingly.
- Hallucinations & provenance gaps: Only adopt tools that show their sources and dates. Train all staff to treat AI outputs as decision support, not automation.
- Compliance risk with scribes: Strictly follow the NHS England guidance. Halt the use of any unregistered tools immediately, as per BMJ coverage and local IG advice.
30-60-90 day adoption plan (department-level)
- Days 1–30: Shortlist your tools. Verify their MHRA registration/class, DTAC claims, and ESF evidence. Define clear KPIs (e.g., time-to-answer, guideline concordance).
- Days 31–60: Run a controlled pilot with opt-in clinicians. Collect metrics on citation click-through and override rates. Rehearse the process for reporting an incident under the new PMS rules.
- Days 61–90: Review the safety and benefit data. Make a clear "scale or stop" decision. Publish a short internal learning report, mirroring the transparency principle of the AI Airlock.
Comparison table
| Platform | Core function | Regulatory status | NHS assurance | Where it’s used/piloted |
|---|---|---|---|---|
| AutoMedica SmartGuideline | AI Guideline Q&A | In MHRA AI Airlock Pilot | Under evaluation | NHS Pilot Sites |
| Prof. Valmed | Generative AI CDS | EU MDR Class IIb (CE) | Requires GB reg & DTAC | EU |
| iatroX | Clinical Reference/Q&A | UKCA Class I, MHRA-registered | DTAC-ready | UK |
| TORTUS | Ambient AI Scribe | MHRA Class I (IIa pending) | DTAC-compliant | NHS Pilot Sites |
FAQ
- What is the MHRA AI Airlock?
- It is a supervised regulatory sandbox for innovative AI as a Medical Device (AIaMD) to be tested and evaluated in real-world NHS settings, with the findings published to inform future rules.
- Did an LLM-based CDS get a higher-risk CE mark?
- Yes, Prof. Valmed has announced that it has achieved EU MDR Class IIb certification for its generative AI clinical decision support tool.
- Is iatroX a registered medical device?
- Yes, first-party pages and the app store listings state that it is a UKCA Class I, MHRA-registered device for informational and educational use.
- Are AI scribes allowed in the NHS?
- Yes, but only if they are registered and governed in line with the official NHS England guidance on ambient scribes.
- What changed in June 2025?
- New, stricter Post-Market Surveillance (PMS) regulations for medical devices in Great Britain took effect on 16 June 2025, requiring manufacturers to more proactively capture and act on real-world safety and performance data.