MHRA AI healthcare rules explained for NHS patients

AI is already doing quiet work inside the NHS. According to the MHRA’s 10 September 2026 announcement, it is being used to help spot strokes and skin cancers earlier and to cut admin through voice tools that free up clinicians’ time. The new question is not whether AI belongs in healthcare at all, but who checks it, how often it is checked and what happens when it gets something wrong. That is why the National Commission into the Regulation of AI in Healthcare matters. Its new report is less about hype and more about rules: how to let useful tools reach patients sooner without treating the public like test subjects. **What this means:** if AI becomes part of your care, safety and accountability are meant to stay in the room with it.

The Commission was set up by the MHRA in September 2025 as an independent advisory body and was led by practising NHS doctors Professor Alastair Denniston and Professor Henrietta Hughes. According to the government announcement, it brought together expertise from health, regulation and AI, including contributors from the USA and Singapore. Over a year, the Commission gathered evidence from more than 12,000 people, which the MHRA says makes this the UK’s biggest public engagement exercise on healthcare technology regulation so far. The Health Foundation, working with Ipsos, also ran workshops with 78 members of the public in Cardiff, Milton Keynes and York between March and April 2026. That matters because the report is not only speaking for officials and tech firms; it is drawing from patients, carers, clinicians and the wider public too.

And the public message was fairly clear. People are open to AI helping with care, but support comes with conditions. The Health Foundation’s research found that accuracy came first, human oversight was non-negotiable, regulation should match the level of risk, and AI should not make care worse for any group. You can hear a simple principle running through all of that: people do not want AI to quietly slip into healthcare without explanation. They want to know when it is being used, they want clinicians to stay responsible for decisions, and they want somebody to be answerable if harm is done. This is not anti-technology. It is basic patient safety.

One of the report’s biggest ideas is staged authorisation for new AI models. The Commission compares it to learner plates: a promising system could be used earlier, but only under close supervision and tight guardrails before it is trusted more widely. In plain English, that means the UK could give patients earlier access to useful tools without pretending early evidence is the same as long-term proof. **What this means:** approval would stop being a simple yes-or-no stamp. A tool could earn greater freedom as it shows, in real settings, that it is safe and effective. For readers who hear ‘faster rollout’ and worry that standards are being lowered, the report is arguing for the opposite idea: move faster where you can, but watch more closely while you do it.

The second big shift is that AI devices would not be judged once and then left alone. The Commission says AI-enabled medical devices should be monitored throughout their working life, because these systems can change after they are deployed. Professor Neil Lawrence, who chaired the technology working group, warned that generative AI does not behave like older medical tools and can produce different results in different circumstances. This point is easy to miss but very important. A scan tool or clinical assistant that works well in testing can perform differently in a busy hospital, with different patient groups or after software changes. Continuous real-world monitoring is meant to catch those changes early, so the MHRA can step in quickly if safety or performance starts to drift.

The report also backs much greater openness with patients. It says the public should be able to search for safety information on specific AI-enabled medical devices, including reports of adverse incidents, building on the MHRA’s existing public safety tools. Alongside that, patients told the Commission they want to know when AI is being used in their care, and the recommendations back a proportionate way of keeping people informed. That may sound like a small administrative fix, but it changes the relationship between patients and technology. **What this means:** instead of AI sitting behind the curtain, there would be a clearer expectation that you can ask what is being used, why it is being used and what is known about its safety record.

The report does not stop at better information. It also says the MHRA needs stronger enforcement powers so it can act decisively when AI systems fall short. That matters because regulation only works if the regulator can do more than advise. If a system is unsafe, inconsistent or poorly monitored, patients need to know there is a route to intervention. Even so, stronger rules on paper will not solve everything. Jennifer Dixon of the Health Foundation said the real test will be whether the NHS has the staff, skills and systems to introduce and monitor AI properly at scale. Other bodies backing the report, including the Health Research Authority, the Professional Standards Authority, the Royal College of Radiologists and pharmacy leaders, made a similar point: safe AI needs training, clear professional expectations and day-to-day oversight, not just good headlines.

What happens next is slower and more political. The government and the MHRA say they will now consider the Commission’s recommendations and publish a formal response later. The full report, its easy-read version and a Welsh-language edition were all released on 10 September 2026, which suggests officials want this debate to reach beyond specialist circles. For the rest of us, the takeaway is straightforward. AI in healthcare is no longer a future-only story; it is already here. The real test is whether the UK can build a system where useful tools reach patients sooner, clinicians stay accountable and the public are not asked to trade trust for speed. If this blueprint is followed, the aim is not simply more AI in the NHS, but better rules around when it is used, who watches it and how patients stay protected.