MHRA Panel Proposes ‘L-Plates’ for AI in Healthcare

An independent commission of NHS doctors has recommended that MHRA authorize new AI-enabled medical software the way the UK licenses learner drivers: in supervised stages, under tight guardrails, before granting the kind of full authorisation a conventional device receives all at once. The National Commission into the Regulation of AI in Healthcare published its report on 10 September 2026, after a year gathering evidence from more than 12,000 patients, clinicians, healthcare leaders, and technology developers.

Why a single approval decision doesn't fit AI

Conventional medical devices are assessed once, against a fixed design, and authorised or refused. The Commission's central argument is that this model does not fit AI systems that can continue to learn and adapt after they reach the market — a static, point-in-time approval captures the device as it existed during review, not as it behaves months or years into real-world use. Its proposed alternative borrows the logic of a UK provisional driving licence: a new AI model would first operate under supervision and tight guardrails, generating real-world evidence of safety and performance, before being granted the kind of fuller, less-supervised authorisation a mature device holds today.

Monitoring doesn't stop at authorisation

The Commission paired the staged-authorisation idea with a recommendation for continuous, real-world monitoring across an AI-enabled device's entire working life, rather than concentrated review activity only at the approval stage. The rationale tracks the same concern: because AI systems can evolve after deployment, regulators need ongoing visibility into how a model is actually performing on real patients, not just how it performed in the evidence submitted for its original authorisation, so that problems that emerge only after wide deployment can be caught and acted on quickly.

A third recommendation would make some of that monitoring visible to the public: a searchable resource where patients and clinicians could look up safety information, including adverse incidents, for a specific AI-enabled medical device. The Commission pointed to MHRA's existing interactive Drug Analysis Profiles — which already let the public search adverse-event data tied to specific medicines — as the kind of tool it has in mind for AI-enabled devices.

What happens next

The Commission's report is a recommendation, not a rule change. Chaired by Professor Alastair Denniston and Professor Henrietta Hughes — both practicing NHS clinicians — and drawing on international contributors from the US and Singapore alongside UK-based experts, its findings now sit with MHRA, which would need to consult on and formally adopt any staged-authorisation framework, continuous-monitoring requirement, or public database before AI-enabled software developers face them as binding requirements. For a device/software regulatory-affairs team already tracking FDA's own moves toward competency-based review of generative AI-enabled devices, the Commission's report is an early look at where the UK's own framework may be headed — not yet a rule to plan against.