Britain's medicines regulator has issued 44 recommendations to overhaul how artificial intelligence products are approved and monitored in the NHS and wider healthcare system, warning that current rules designed for traditional medical devices are inadequate for rapidly evolving AI technologies.
The Medicines and Healthcare Products Regulatory Agency (MHRA), which oversees all medical devices and licenses drugs in the UK, published its recommendations as AI use in healthcare is set to become routine. The agency established the National Commission on the Regulation of AI in Healthcare in September 2025 as an independent advisory body to develop a framework balancing innovation with patient safety.
Lawrence Tallon, MHRA chief, told the BBC that patients will increasingly encounter AI as part of routine NHS care.
"What I would expect is that patients will increasingly see AI as part of the way that normal NHS healthcare is delivered. That should happen in a way that they can maintain their trust and their confidence in what's happening."
The recommendations emerged from an independent commission that gathered input from more than 12,000 people, including patients, clinicians, and healthcare professionals. The MHRA published supporting evidence on 11 June 2026 from its engagement programme to inform the future regulatory approach.
What are the key recommendations?
The commission's proposals address fundamental gaps in how AI systems should be regulated differently from conventional medical devices. The recommendations include establishing continuous monitoring of AI products after approval, with powers to remove them from regulatory clearance if they malfunction or become less effective over time. This reflects a core challenge with AI systems: unlike static medical devices, they continue to evolve after deployment as they process new data and adapt their behaviour.
The proposals also call for patients to have explicit rights to know whether AI is involved in their care and to access clear information about the specific products being used. Additionally, regulators should gain the power to penalise developers whose AI products fail to meet required standards.
A further recommendation proposes an AI "L plate" system—similar to learner plates on vehicles—that would allow new AI models to be trialled by healthcare professionals under close supervision before receiving fuller regulatory approval. An advisory-panel report published on 9 September 2026 urged this staged approval process to balance the need for innovation with patient protection.
Why do current rules not work for AI?
Tallon explained that the existing medical devices regulatory framework dates from an era focused on static products such as hip replacements, knee replacements, stethoscopes and plasters. While these rules may suffice for simple AI applications trained to identify known symptoms on medical scans, they fail to address the behaviour of more sophisticated models.
"Unlike most of the medical products we're used to regulating, these products continue to change after the point of authorization. As new data gets fed in, they learn, they adapt, they drift."
This drift—the gradual shift in how an AI system performs as it encounters new information—poses a regulatory puzzle that traditional frameworks cannot solve. Tallon acknowledged that regulating AI remains a global challenge with no country yet having definitively solved the problem.
How are AI scribes being used in general practice?
AI note-taking systems powered by large language models, known as scribes, are already in widespread use. Approximately 40% of UK-based GPs employ these tools to record consultations and generate clinical reports, automating administrative work that traditionally consumed significant clinician time.
However, research raises concerns about patient willingness to disclose sensitive information when AI is involved. A study from the University of Edinburgh found that patients were less likely to share personal details such as substance abuse history if they knew the conversation was being processed by AI.
Professor Henrietta Hughes, a GP who contributed to the commission's work, noted that while many of her patients accepted AI use during consultations, some chose to opt out.
"Some say, 'I don't want to talk to a robot', and that is also fine."Hughes acknowledged that AI scribes can make errors in their notes but stressed that doctors bear responsibility for reviewing and correcting them.
Concerns about AI scribes extend beyond patient privacy. Healthwatch warned on 31 August 2026 that AI scribes could get drug names and diagnoses wrong, and noted that the MHRA had not classified them as medical devices, leaving them outside the current regulatory framework.
What opportunities does AI present for healthcare?
Despite regulatory challenges, healthcare leaders see transformative potential in AI. Currently, the technology is primarily used to assist with administrative tasks or carry out symptom spotting under human expert supervision. However, enthusiasm for AI's broader applications in health and medical research remains high.
Professor Alastair Denniston, an ophthalmologist who worked on the commission, described AI as
"an exceptional opportunity for healthcare which was likely to rank alongside step-changes such as antibiotics and MRI."Such comparisons underscore the scale of change many in the sector anticipate.
Yet this optimism must be tempered by real risks. Concerns persist about AI models making incorrect decisions due to biases embedded in the patient data used to train them. Additionally, AI chatbots have been known to provide medically inaccurate advice, raising questions about how such systems should be deployed in clinical settings.
What happens next?
The commission's final recommendations were expected in late summer 2026, with the MHRA set to use them to shape future regulatory rules. The recommendations will feed into a new framework for AI in healthcare that aligns with the 10-Year Health Plan for England and the Life Sciences Sector Plan. The MHRA's commission page was updated on 9 September 2026, suggesting that formal publication or policy response was imminent as the regulatory landscape for AI in healthcare continues to take shape.
