Executive overview
In 2025, the UK's regulatory landscape for clinical artificial intelligence is rapidly maturing. The MHRA AI Airlock, a regulatory sandbox for AI as a Medical Device (AIaMD), is moving from its initial pilot to an expanded Phase II, with its findings set to shape future rules and accelerate safe access to innovation (GOV.UK).
In parallel, new Post-Market Surveillance (PMS) regulations, which came into force on 16 June 2025, have tightened the duties for all medical device manufacturers—including software—to continuously monitor real-world performance and report incidents. For UK clinicians and healthcare leaders, this is a pivotal moment. The processes of procuring, piloting, and using AI tools now require robust monitoring and evidence governance, placing clinicians at the centre of providing that essential real-world evidence (GOV.UK, NICE).
AI Airlock in one page (what it is / isn’t)
The MHRA AI Airlock is a supervised, time-boxed sandbox designed for cutting-edge AIaMD. Its purpose is to allow for the real-world testing of these devices in a controlled NHS environment, helping to identify regulatory gaps, develop appropriate mitigations, and generate evidence before a wide-scale rollout. The outputs include public reports on each product's performance and key learnings that are shared across regulators (GOV.UK).
Launched in Spring 2024 with an initial pilot of five technologies, Phase II was funded and expanded in 2025, with a new cohort of nine technologies currently undergoing selection. The Airlock is a key part of the MHRA’s broader device reform roadmap, developed in collaboration with NHS England, NICE, and Approved Bodies (GOV.UK, medregs.blog.gov.uk).
How Airlock interacts with NICE & NHSE routes
The AI Airlock does not exist in isolation. It is part of a broader, increasingly joined-up pathway for digital health adoption in the UK.
- NICE Evidence Standards Framework (ESF): The ESF is the foundational yardstick for any digital or AI tool. It sets out the expectations for clinical and economic evidence that a manufacturer must provide to prove their product's value. Clinicians and commissioners should use this to assess any new tool (NICE).
- AI & Digital Regulations Service: This multi-agency service, led by NICE, acts as a "front door" for innovators and NHS teams, helping them navigate the complex approvals and evaluation landscape, from MHRA registration to NICE guidance. The clear direction of travel is towards a more streamlined, collaborative process across the MHRA, NICE, and NHS England, designed to shorten the path for safe and effective AI and digital health tools to reach the frontline (Taylor Wessing).
What the new PMS rules mean on the ward/clinic floor
The new Post-Market Surveillance regulations, effective from 16 June 2025, have significant practical implications. The strengthened duties mean that manufacturers must now continuously and proactively gather safety and performance data on their devices once they are on the market. They must analyse trends and have clear pathways for reporting incidents to the MHRA (GOV.UK).
For clinicians, this means your feedback is now a formal, integral part of a device's regulatory compliance. Your incident reports, observations on performance, and usage data are essential inputs for the manufacturer's mandatory PMS plan. Expect to see clearer prompts for feedback within products and more structured processes for reporting adverse incidents.
Clinician roles across the lifecycle
- Pre-market (Airlock pilots): Clinicians at pilot sites can co-design the clinical tasks for the AI, help define the key performance indicators (KPIs) like guideline concordance and error intercepts, and provide crucial real-world feedback.
- Go-live: Before adopting any new tool, clinicians should check the vendor’s evidence against the NICE ESF, demand transparency on data sources (citations and last-updated dates), and confirm that the Trust has a full Data Protection Impact Assessment (DPIA) and clinical safety case in place.
- Post-market: Under the new PMS rules, clinicians have a vital role in recording and reporting adverse incidents and "near-misses," and contributing to the periodic safety update evidence that is now mandatory for manufacturers.
Where this lands for common AI categories
- Decision support / retrieval-augmented answers: These tools are permitted when they are clearly positioned as support, not automation. They must provide citations and require human verification. This makes them ideal for educational and reference use cases, such as iatroX-style cited Q&A.
- Ambient voice / AI scribes: These tools can only be adopted under strict organisational approval, with a full clinical safety case and human verification of every output. Local Standard Operating Procedures must align with the new PMS requirements for data capture and incident reporting.
30–60–90 day plan for a department trial (Airlock-ready habits)
- 30 days: Pick one low-risk, high-volume task. Define objective KPIs and create a simple incident taxonomy. Verify the vendor’s NICE ESF mapping and ask to see their formal PMS plan.
- 60 days: Run a controlled pilot with a small cohort of opt-in clinicians. Collect data on metrics like citation-click-through and acceptance/override rates. Rehearse the process for reporting a simulated incident to the MHRA.
- 90 days: Review the safety and benefit data from your pilot. Make a clear "scale or stop" decision. Publish a short internal learning report, mirroring the transparency principle of the AI Airlock.
Buyer’s checklist (for CCIOs/clinical leads)
- Regulatory status: Is the tool an AIaMD? If so, what is its intended purpose and current regulatory pathway (e.g., standard UKCA marking, currently in the AI Airlock)?
- Evidence: Which NICE ESF tier does it meet? Is there an external evaluation or a plan for real-world evidence generation?
- PMS readiness: Does the vendor have a clear PMS plan? Are there in-product hooks for reporting? Do they commit to sharing safety signals with the Trust?
- Governance: Is a DPIA complete? Is there a clinical safety case and a clear audit trail? Is there a defined route for handling field safety actions or recalls?
Measurement that matters
- Safety/quality: Incident and near-miss rates; time taken to implement corrective actions.
- Effectiveness: Time-to-answer; time-to-note; concordance with clinical guidelines.
- Adoption: Suggestion acceptance rates; override rates with documented justifications.
- Equity/robustness: Performance stratified by different patient groups; monitoring for "model drift" after guideline updates.
FAQs
- What is the MHRA AI Airlock?
- It is a supervised, time-bound regulatory sandbox that allows innovative AI as a Medical Device (AIaMD) to be tested and evaluated in real NHS settings, with the findings published to inform future regulation.
- Is Phase II of the Airlock live?
- Funding for Phase II was announced in early 2025. Applications have closed, and the selection of the nine new technologies for the cohort is currently in progress, with updates expected in Summer 2025.
- What changed on 16 June 2025?
- New, strengthened Post-Market Surveillance (PMS) regulations for all medical devices, including software, took effect. These place stronger duties on manufacturers for ongoing vigilance and real-world performance monitoring.
- Where do NICE/NHSE fit?
- The NICE ESF guides the evidence required for a tool. The AI & Digital Regulations Service (led by NICE and involving NHS England) helps both innovators and NHS teams navigate the complex adoption pathway.
Closing (what to do next)
The MHRA AI Airlock provides a clear preview of what future-proof clinical AI practice will look like: transparent pilots, a focus on measurable benefits, and rigorous post-market learning. The best way to prepare is to build these habits now. By embedding the principles of cite, verify, and report into your daily use of any digital tool, your team will already be operating to the new standard when the next generation of AIaMD arrives in your clinical pathway.