AI Oversight Review Urges Standards Comparable to Medical Licensing

A government commission examining how artificial intelligence should be regulated in healthcare has proposed that the most autonomous AI systems be assessed under a framework that mirrors the standards applied to doctors, including ongoing scrutiny rather than one-off approval.

The National Commission into the Regulation of AI in Healthcare, launched by the Medicines and Healthcare products Regulatory Agency in September 2025, brings together AI specialists, clinicians and regulators to advise on a new regulatory framework, with recommendations due to be published later this summer. The Commission's work addresses a fundamental gap in current oversight arrangements, which were designed for static software rather than systems capable of autonomous judgement.

The central argument put forward by the Commission is that existing medical device regulation is poorly matched to what has become known as agentic AI. Such systems can take in goals, decide for themselves what steps to take, and then act to carry out those steps without constant human instruction. The more independently a system behaves, the less able human supervisors are to anticipate or prevent adverse outcomes — a problem that direct analogies with software certification have so far failed to address.

Under current rules, low-risk devices can be self-certified by the manufacturer, while medium and high-risk devices are subject to independent conformity assessments before they are placed on the market. The Commission's emerging position is that this pre-market approval model, which in effect treats AI as a finished product at the point of sign-off, does not adequately account for systems that continue to learn and adapt once deployed.

The proposed alternative draws directly from the structure of clinical professional standards. The assessment of a clinician's fitness to practise does not end at qualification; it is subject to revalidation, performance review and the prospect of sanction if standards fall. The Commission has indicated that highly autonomous AI systems should be held to an equivalent principle, with ongoing monitoring replacing the current assumption that approved software remains safe indefinitely.

Questions of liability also feature prominently in the Commission's scope, including how accountability should be distributed when an AI system acts autonomously and produces an adverse outcome. The so-called "black box" problem complicates this considerably: because AI systems do not process information in natural human language, clinicians may be unable to explain why a system produced a particular recommendation, let alone demonstrate that they should have overridden it. Placing full liability on healthcare professionals in such circumstances has been questioned as neither fair nor workable.

The MHRA published in June 2026 the findings of its extensive call for evidence and public engagement programme, drawing on submissions from 760 individuals and institutions as well as research conducted with patients, frontline health staff and technology experts across the UK. Those findings are now being used to finalise the Commission's recommendations.

Separately, the MHRA announced in June 2026 the creation of a new regulatory sandbox that will allow AI tools to be tested in controlled conditions alongside regulators, beginning with industry and academic partners from summer 2026. The sandbox is intended to build the evidence base required to support faster and safer deployment of AI across the health service.

The Commission's work sits within the government's stated ambition to make the NHS the most AI-enabled health system in the world, as set out in its ten-year health plan. Whether the medical licensing analogy translates cleanly into statutory instruments remains to be seen. The MHRA chief executive Lawrence Tallon has committed that the agency will act on the Commission's recommendations to support the NHS's digital transformation. The regulatory rulebook is expected before the end of the year.