MHRA Calls for 44 New AI Rules in NHS Healthcare — SkimNews
Get the Health newsletter
Daily health & science — research, biotech, public health, the studies worth knowing. Free.
- MHRA published 44 recommendations to regulate AI in healthcare, compiled by an independent commission with input from more than 12,000 patients and clinicians
- Key proposals include continuous monitoring of AI products with removal from approval if they malfunction, a patient right to know when AI is involved in their care, developer penalties for substandard products, and an 'L plate' system for supervised trials of new AI models
- MHRA chief Lawrence Tallon said the existing medical devices framework dates from an era of hip replacements and stethoscopes and does not fit AI products that 'learn, adapt, drift' after the point of authorization, adding that no single country has 'absolutely cracked' AI regulation
- AI note-taking tools known as scribes, driven by large language models, are reportedly used by 40% of UK-based GPs to record consultations and generate reports
- A University of Edinburgh study found patients were less likely to share personal information such as substance abuse history when they knew AI was processing the conversation
- Commission member Professor Henrietta Hughes, a GP, said some patients opt out of AI-assisted consultations, telling her 'I don't want to talk to a robot'
- Professor Alastair Denniston, an ophthalmologist on the commission, called AI an 'exceptional opportunity' likely to rank alongside antibiotics and MRI, though concerns remain about biased training data and incorrect chatbot medical advice
Why it matters: The recommendations arrive with AI scribes already in 40% of UK GP surgeries, yet the MHRA concedes no country has 'cracked' AI regulation. For the NHS's 12,000+ consulted patients and clinicians, the concrete shifts are a new right to be told when AI is involved in their care and a regime that can pull malfunctioning products from the market — moving the UK from a framework built for static devices toward one that can govern self-learning systems.
Ask SkimNews
