New AI tools entering the NHS would first run under what the UK’s regulators are calling L-plates — a probationary licence with close supervision and tight guardrails, before any full approval. The recommendation comes from the National Commission into the Regulation of AI in Healthcare, which published its blueprint on 10 September 2026 after a year gathering evidence from more than 12,000 people.
The Commission was set up by the Medicines and Healthcare products Regulatory Agency (MHRA) in September 2025 and chaired by two practising NHS doctors, professors Alastair Denniston and Henrietta Hughes. Its full recommendations are published on GOV.UK, and the Financial Times gave the L-plates framing its headline — provisional approval for new AI healthcare tools until manufacturers can demonstrate they are safe in the wild.
Why medicine treats software differently now
The old model of medical device regulation assumes a product is finished before it ships. A stent either meets spec or it doesn’t, and the spec doesn’t change overnight. AI models break that assumption because they keep changing after deployment — retrained, updated, or quietly degrading as the data around them shifts.
The Commission’s answer is a two-part shift. First, staged authorisations: new models deploy under supervision and demonstrate real-world safety and performance before earning fuller authorisation. Second, continuous real-world monitoring across the device’s whole working life, replacing the single point-in-time approval entirely. As the Commission’s own summary puts it, this “reflects how AI technologies evolve after they are deployed, rather than at the point of approval alone.”
The MHRA’s chief executive Lawrence Tallon welcomed the report and said the opportunities — faster care, more clinician time with patients — would only be realised “if we have a modern, dynamic regulatory framework that accelerates safe adoption, protects patients and commands public and professional confidence.” The government and MHRA will now consider the recommendations, with a formal response to follow. Nothing is law yet.
What the public actually asked for
The evidence base here is unusually large for a regulatory report. Over a year, the Commission took submissions from patients, clinicians, industry and the public — more than 12,000 people, which the government calls the largest engagement of its kind ever undertaken in the UK on healthcare technology regulation. Deliberative research by the Health Foundation, published alongside the report, filled in the detail.
The pattern in that evidence was consistent: people are broadly open to AI in their care, with three conditions. Strong safety standards. Meaningful human oversight. And transparency — patients want to know when AI is being used on them. The Commission also recommends a public search tool for the safety record of specific AI-enabled medical devices, including adverse incidents, modelled on the MHRA’s existing drug safety profiles.
That last recommendation is the one with the sharpest teeth. A searchable public record of AI device failures changes the incentive structure for manufacturers in a way that guidance documents never will.
The New Zealand angle
New Zealand has no equivalent framework, and the gap is not theoretical. Health NZ staff were caught using unapproved chatbots for clinical notes in May — the shadow-AI problem thrives exactly where formal approval pathways are missing or too slow. The NHS app’s AI triage rollout for 200,000 patients drew warnings from health leaders that the evidence base was thin. Both stories are versions of the same missing piece: a middle ground between “banned” and “deployed,” where new tools run under supervision while the evidence accumulates.
Whether Medsafe — which regulates therapeutic products in New Zealand — could run an L-plates scheme with current staffing and powers is an open question. The MHRA is one of the world’s better-resourced medical regulators, and the Commission still felt the need to recommend enhanced enforcement powers alongside the staged approvals.
There is also a timing question worth asking. The UK spent a year and 12,000 consultations getting here. AI diagnostic tools are already in use in NZ hospitals — AI is already detecting strokes and skin cancers in the NHS under existing rules, and the same is true here in smaller pilots. The regulation is arriving, in every country, after the tools it governs.
FAQ
What does “L-plates” mean for AI health tools? Staged authorisation: a new AI model deploys under close supervision and tight guardrails, and must demonstrate real-world safety before receiving fuller approval — like learner plates for a driver.
Who produced the UK recommendations? The National Commission into the Regulation of AI in Healthcare, established by the MHRA in September 2025 and chaired by NHS consultants Alastair Denniston and Henrietta Hughes. It published its blueprint on 10 September 2026.
Are these rules in force yet? No. The Commission’s recommendations now go to the UK government and MHRA for a formal response. Nothing has been implemented.
Does New Zealand regulate AI in healthcare? Not with a dedicated framework. Medsafe handles therapeutic products, but there is no NZ equivalent of the staged approval and continuous monitoring model the UK Commission is proposing.
Related reading: Health NZ Staff Caught Using ChatGPT for Clinical Notes, The NHS Just Put AI Triage in 200,000 Patients’ Pockets, and AI Regulation: Australia vs NZ.
— CJ Murden, editor of Singularity.Kiwi. Former digital technologies teacher, author of AI-focused books. Writing with a New Zealand focus.
