In the UK, AI medical devices fall under the Medical Devices Regulations 2002 (UK MDR 2002). The Medicines and Healthcare products Regulatory Agency (MHRA) enforces these regulations. Official MHRA and Government publications describe no separate AI-only regime in force today. So for now the rule is straightforward. AI software is regulated as a medical device whenever its intended purpose is medical.
This regulatory position may change as the Government moves forward with its plans for AI in healthcare. On 6 October 2026, the Government accepted all 44 recommendations of the National Commission into the Regulation of AI in Healthcare. It’s a significant step. But most of this work is still proposed or in development, so it is not yet law.
The sections that follow explain the current requirements, the reforms planned for 2026–2027, the MHRA AI Airlock and what manufacturers can do to prepare.
What Are AI Medical Devices in the UK?
An AI medical device is AI software that meets the legal definition of a medical device. This category is known as AI as a Medical Device (AIaMD). It sits within a wider group called Software as a Medical Device (SaMD). This covers software that counts as a device on its own, with no hardware attached.
Intended purpose decides everything. MHRA guidance says a product’s medical purpose is determined by its labelling, its instructions for use and its promotional material, app store descriptions included. Even a “for information only” disclaimer won’t change the outcome if the product makes medical claims elsewhere.
Not all healthcare AI is a medical device. The MHRA’s standalone software guidance gives these examples:
- Unlikely to be a device: Software that only books appointments, an electronic record that simply replaces paper, or reference information a clinician uses to decide.
- Likely to be a device: Software that uses an algorithm to influence dose, support diagnosis, or give a prognosis or an individual disease risk.
AI can sit inside standalone software, imaging systems, monitoring systems or decision-support tools. In complex systems, only the module that meets the definition may need regulating. UK law has no separate AI classification. Existing classification rules apply.
How Are AI Medical Devices Regulated in the UK?
Under the current system, AI medical devices follow the same conformity route as other devices. UK MDR 2002 sets that route, and the MHRA enforces it. MHRA guidance sets out the rules for Great Britain (GB: England, Scotland and Wales) and Northern Ireland (NI).
Great Britain
- Manufacturers based outside the UK must appoint a UK Responsible Person.
- Devices carry either a UKCA (UK Conformity Assessed) mark or a CE mark that is still accepted.
- CE-marked devices stay accepted until 30 June 2028 or 30 June 2030, depending on the legislation they comply with. The same GOV.UK page says the Government plans to consult on indefinite recognition of CE marking.
- Self-declaration is allowed for Class I devices that aren’t sterile or measuring. Other classes need a UK Approved Body.
Northern Ireland
- The EU Medical Devices Regulation (EU MDR) and In Vitro Diagnostic Regulation (EU IVDR) apply in Northern Ireland under the Northern Ireland Regulations 2021.
- CE marking is required. UKCA isn’t recognised.
- Manufacturers based in GB need an EU or NI authorised representative.
- The MHRA remains the competent authority for incident reporting. See the NI guidance.
Technical documentation, clinical evidence and post-market surveillance work the same way for AI devices as for other devices.
What Are the UK AI Medical Device Requirements?
UK AI medical device requirements are the existing UK MDR 2002 requirements applied to software.
- Intended purpose and classification: What you claim defines your purpose, and your purpose determines your classification. Software that drives or influences another device takes that device’s class. That classification then decides whether you self-declare or go through an approved body.
- Risk management: Risks have to be acceptable against the benefits. Cut them as far as you can, and tell users about whatever risk remains. The MHRA lists ISO 14971 as a useful standard. Following it can help you show compliance, but the legal duty itself comes from UK MDR 2002
- Clinical evaluation and performance: Conformity must include a clinical evaluation, and you need evidence that the software does what you claim. Novel software may call for clinical investigations. Not every AI device needs a trial.
- Software lifecycle: The software must be validated to the state of the art. That means considering lifecycle, risk management, verification and validation. Version control, change control and maintenance usually sit within the manufacturer’s quality system, alongside lifecycle standards such as IEC 62304.
- Technical documentation: It must show conformity with the essential requirements. The UK Responsible Person keeps it, along with the declaration of conformity, available to the MHRA.
- Cybersecurity: No MHRA AI-specific cybersecurity checklist is currently in force. Data protection law still applies, and the Government has said draft cybersecurity guidance is on the way.
- Post-market surveillance (PMS): Keep monitoring the device in use so problems are caught, reported and acted on. For AI products, that means tracking real-world performance. The MHRA suggests a registration or activation system for apps, so affected users can be traced during corrective action. Strengthened PMS rules for GB came into force on 16 June 2025.
- Vigilance: Manufacturers report incidents through the MHRA’s vigilance system and take corrective action. Clinicians and patients report via Yellow Card.
Why Is UK AI Medical Device Regulation Changing?
Today’s rules lean on assessment at one point in time, and AI can change after release. The Government’s own response says UK MDR 2002 predates today’s AI. It also says the regulations aren’t fit for purpose for AI.
The practical problems are these:
- Adaptive AI can learn after release.
- Model updates can change behaviour.
- Drift means performance shifts as real-world data changes.
- Bias can lead to weaker results for some groups.
- A single pre-market review can’t capture any of this.
Current controls are UK MDR 2002 plus PMS and vigilance. The measures below are proposed responses. The response describes a shift towards ongoing assurance instead of heavy reliance on pre-market assessment and passive incident reporting.
What Is Changing in UK AI Medical Device Regulation in 2026-2027?
The Commission published its report on 10 September 2026, and the Government accepted all 44 recommendations on 6 October. This is the main 2026 update to UK medical device regulation for AI. Accepted means the Government will take them forward. It does not mean they are law.
| Measure | Status on 8 Oct 2026 |
| Qualification and classification (Rec 1) |
Accepted; consultation not yet launched. Not yet law
|
| Change control plans (Rec 6) |
Accepted; guidance in development. Not yet law
|
| Tailored post-market surveillance (Rec 17) |
Accepted; legislation may be needed. Not yet law
|
| Risk-proportionate oversight (Recs 3-5) |
Accepted; options being considered. Not yet law
|
| Staged authorisation (Rec 14) |
Accepted; MHRA exploring options. Not yet law
|
| Transparency and equity (Recs 9, 13, 37) |
Accepted; guidance to follow. Not yet law
|
AI Medical Device Classification Changes
Proposed change: The MHRA will consult, then use secondary legislation to introduce three things:
- A revised device definition, which may carve out certain administrative, wellbeing and decision-support software.
- A classification approach that reflects the risks and benefits of software and AI more closely.
- A definition of intended purpose that takes design and functionality into account.
Until then, the MHRA plans guidance updates under the existing rules. The Commission also asked the MHRA to address limits of self-declared Class I devices. This is not yet a current legal requirement.
The MHRA has also put draft Medical Devices (Amendment) Regulations 2026 out for comment, published on 8 May 2026. Its announcement names compulsory UDI (Unique Device Identification), international reliance and claims alignment. Software-specific classification changes aren’t mentioned, and there is no commencement date.
AI Medical Device Change Management
Proposed change: By December 2026, the MHRA intends to issue draft guidance on predetermined change control plans (PCCPs). These plans would let manufacturers define in advance which changes are allowed, including changes to adaptive AI, provided the intended use remains the same.
No mandatory change-management procedure is in place yet, and the guidance is not yet published.
Post-Market Surveillance for AI Medical Devices
Current requirement: Existing PMS and vigilance duties, strengthened in June 2025.
Proposed change: The MHRA will build on these with a toolkit of measures that could be required individually or in combination. The Commission’s menu includes:
- PMS plans assessed at authorisation.
- Real-world post-market studies.
- More regular performance reporting.
- Escalation when performance degrades, even without a reportable incident.
Related items cover Yellow Card improvements, a public adverse-incident search tool and guidance on monitoring performance across population groups. Legislative reform may be needed.
Risk-Based AI Medical Device Regulation
Government commitment: The MHRA will consider adding safety requirements where evidence shows a need. It will consider limiting oversight of very low-risk software, for example through enforcement discretion, and regulation focused on the medical function of multi-purpose products.
No final AI-specific risk classification is legally binding today.
Staged Authorisation for AI Medical Devices
Government commitment: The MHRA will explore staged authorisation. That could draw on proposed licensing powers and statutory sandbox legislation, with a clear route from staged to full authorisation It is an accepted recommendation under exploration. It is not an operational MHRA approval route.
AI Medical Device Transparency and Health Equity
Proposed change: The MHRA plans guidance on model cards, user-centred design and dynamic labelling. It also plans guidance on showing that devices don’t create or widen disparities for groups such as children, women and ethnic minorities, covering representative data and performance across groups. It will work to publish a regularly updated list of authorised AI-enabled devices.
The Government response sets no explainability requirement. These items are at guidance stage.
What Is the MHRA AI Airlock?
The MHRA AI Airlock is a regulatory sandbox for AIaMD. Innovators and regulators work with real products to explore regulatory challenges, and the test results become evidence for future MHRA guidance and policy. That is how the programme supports AI medical device regulation in the UK.
Phase 3 focuses on post-market surveillance and lifecycle oversight. Key facts:
- The MHRA began accepting applications on 6 October 2026The deadline for the current review wave is 31 October, but applications will remain open after that.
- The first wave is due to be selected in November 2026, with an applicant webinar on 22 October.
- There is no fee, but applicants fund their own testing. Plans usually run 6-12 months, and funding runs to April 2029.
- The three test settings are simulation, virtual and real-world. Real-world testing runs outside live patient care, for example as shadow deployment.
Taking part in the AI Airlock does not grant regulatory approval or replace the need for MHRA authorisation, registration, UKCA certification or conformity assessment. Applicants should share their regulatory documentation where available.
Pay attention if your device raises post-market questions, because your project could shape future guidance.
UK AI Medical Device Regulation Timeline 2026-2027
| Date | Milestone | Status |
| 8 May 2026 | Draft Medical Devices (Amendment) Regulations 2026 published for comment (separate from the Commission’s recommendations) |
Draft; no official commencement date
|
| 10 Sep 2026 | Commission publishes its 44 recommendations | Completed |
| 6 Oct 2026 | Government accepts all 44; AI Airlock Phase 3 opens | Completed |
| 31 Oct 2026 | Cut-off for the current Airlock review wave (applications stay open after) | Official date |
| Nov 2026 | First Airlock Phase 3 cohort selected | Planned |
| Dec 2026 | Draft guidance on predetermined change control plans | MHRA intends |
| Spring 2027 | Classification and qualification consultation | MHRA intends |
| Spring 2027 | Implementation roadmap and further details on most recommendations |
Government commitment
|
No official publication yet confirms when secondary legislation will be laid or take effect.
How Can Manufacturers Prepare for UK AI Medical Device Regulations?
These are practical readiness steps. Start with product qualification: test your claims, labelling and app store text against the MHRA’s guidance. Then review classification, checking it against today’s rules while you watch for the Spring 2027 consultation.
Your clinical evidence needs the same scrutiny. Confirm it covers your real user groups and clinical settings. For AI products, strengthen change management by documenting your controls for model updates, software updates, performance changes, validation and version control. Post-market monitoring deserves attention too. Track performance, complaints, incidents and model behaviour, and measure performance across populations.
Your technical documentation should reflect your intended purpose, risk management, architecture, verification and validation, clinical evidence, cybersecurity and PMS. Check each of those. Keep an eye on MHRA developments as well, since draft guidance and consultations will appear first on GOV.UK. Finally, prepare for 2027 changes by building flexibility into your quality system now. Don’t wait for the final rules.
Conclusion
UK AI medical device compliance today means meeting UK MDR 2002. The 2026-2027 changes are announced, but most of them have not yet been legislated. Readiness means keeping four things apart. These are the requirements in force, the reforms already announced, the proposals still being developed and the future implementation rules.
Artixio supports manufacturers with AI medical device qualification, classification, technical documentation, PMS planning and regulatory strategy. Explore our medical device regulatory consulting services for support across the product lifecycle.
FAQs
Q1. How are AI medical devices regulated in the UK?
Through UK MDR 2002, enforced by the MHRA, when the software has a medical purpose. There’s no separate AI law in force. AI as a medical device in the UK follows the same classification, conformity, evidence and PMS rules as other devices.
Q2. What are the UK AI medical device requirements?
Registration, a correct class, conformity assessment where required, risk management, clinical evaluation, validated software, technical documentation, PMS and vigilance. Cybersecurity and data protection also apply.
Q3. What is changing in UK AI medical device regulation in 2026?
The Government accepted all 44 Commission recommendations on 6 October 2026. AI Airlock Phase 3 opened, and the MHRA intends draft change-control guidance by December 2026. Nothing here is in force yet.
Q4. What is the MHRA AI Airlock?
A regulatory sandbox for AIaMD UK. Phase 3 tests PMS and lifecycle oversight, and it doesn’t replace approval.
