- The MHRA has proposed 44 recommendations for regulating AI products used in healthcare.
- AI systems can change after approval, creating a challenge for rules designed around traditional medical devices.
- Patients could gain a right to know when AI is involved in their care and how the technology is being used.
Artificial intelligence is moving from the edges of the NHS into everyday healthcare, but Britain’s medical regulator says the rules governing it were largely designed for a very different kind of technology.
The Medicines and Healthcare products Regulatory Agency (MHRA) has published 44 recommendations for updating the regulatory framework as AI becomes more widely used in healthcare.
The proposals were developed by an independent commission that consulted more than 12,000 people, including patients and clinicians.
MHRA chief executive Lawrence Tallon said AI would soon become a normal part of NHS healthcare, but that its growing presence needed to happen without undermining patients’ confidence.
“What I would expect is that patients will... increasingly see AI as part of the way that normal NHS healthcare is delivered,” Tallon told the BBC.
“That should happen in a way that they can maintain their trust and their confidence in what’s happening.”
The recommendations include continuous monitoring of AI products after approval, the ability to remove products from regulatory approval if they stop working effectively, and powers to penalise developers when products fail to meet required standards.
Patients could also have a right to know when AI is involved in their care and access information about the products being used.
Why AI is different from a stethoscope
The biggest problem for regulators is that an AI medical product may not remain exactly the same after it has been approved.
Tallon said the existing medical devices framework was developed largely around products such as 'hip replacements and knee replacements' as well as 'stethoscopes and plasters.'
That model becomes harder to apply to AI systems that can be updated or change as they encounter new data.
“Unlike most of the medical products we're used to regulating, these products continue to change after the point of authorization,” Tallon said.
“As new data gets fed in, they learn, they adapt, they drift.”
That creates a different regulatory challenge. A traditional medical device can be tested and assessed before it reaches patients, whereas an AI system may behave differently after deployment.
The MHRA therefore wants greater powers to monitor products continuously rather than treating approval as the end of the regulatory process.
One proposal is an AI “L plate” system, allowing new AI models to be trialled by healthcare professionals under close supervision before they are more widely deployed.
The proposals come as AI is already being used in parts of the NHS. AI-powered medical scribes, which listen to consultations and generate clinical notes, are reportedly being used by around 40 per cent of UK GPs.
The trust problem goes beyond regulation
The technology may save doctors time, but its use also raises questions about what patients are willing to share when they know an AI system is involved.
A University of Edinburgh study found patients could be less likely to disclose sensitive information, including substance abuse histories, if they knew their consultation was being processed by AI.
That could become important if AI moves beyond administrative tasks and plays a greater role in diagnosis and treatment.
Professor Henrietta Hughes, a GP who worked on the MHRA report, said many of her patients were comfortable with AI being used during consultations, but some chose not to participate.
“Some say, ‘I don’t want to talk to a robot’, and that is also fine,” she said.
AI scribes can also make mistakes, although Hughes said doctors remain responsible for checking and correcting their notes.
That human oversight is currently one of the main safeguards around medical AI. But as systems become more capable, regulators face the question of how much responsibility can realistically remain with a doctor when an AI system is involved in a clinical decision.
The potential upside is enormous. Professor Alastair Denniston, an ophthalmologist who worked on the commission, called AI “an exceptional opportunity” for healthcare and said it was “likely to rank alongside step-changes such as antibiotics and MRI”.
The technology could also transform medical research. Arm chief executive Rene Haas has predicted that AI could produce a cure for cancer “within our lifetime”.
But the promise comes with risks ranging from biased training data to incorrect medical advice and AI systems producing confident but wrong conclusions.
The MHRA's recommendations are therefore an attempt to address a problem that traditional medical regulation was never designed to solve.
Britain is not trying to stop AI from entering healthcare. It is trying to build a system that can keep watching it after it gets there.
And that may prove to be the defining challenge of regulating AI in medicine: not deciding whether a system is safe once, but determining whether it remains safe as the technology continues to learn, change and evolve.
