The UK’s medical watchdog has warned that national laws must be fundamentally rewritten to protect patients as “adaptive” artificial intelligence becomes an everyday part of NHS care.
The National Commission into the Regulation of AI in Healthcare published a landmark report on 10 September 2026, setting out 44 recommendations to modernize the way the health service handles technology that can learn and change after it is deployed.
Lawrence Tallon, Chief Executive of the Medicines and Healthcare products Regulatory Agency (MHRA), said that existing regulations are no longer fit for purpose. He noted that current laws were largely designed for a previous era of “static” medical devices, such as stethoscopes, hip replacements, and plasters, which do not change their behaviour once they leave the factory.
In contrast, modern AI models used in diagnosis and treatment can “drift” or adapt as they process more patient data, potentially leading to different outcomes over time.
The ‘L-plate’ system for AI
One of the core proposals in the GOV.UK report is the introduction of a staged authorisation process, described as an “L-plate” system for AI models.
Under this framework, new AI tools would be granted restricted, provisional approval while safety data is gathered in real-world clinical settings. This would allow the MHRA to monitor a product’s performance continuously rather than granting a single, permanent point of approval. If a tool’s effectiveness drops or it begins to “drift” from its original safety standards, regulators would have the power to strip it of its authorisation or penalise the developers.
The commission, led by NHS doctors Professor Alastair Denniston and Professor Henrietta Hughes, also emphasised the need for transparency. A key recommendation is a formal “right to know” for patients, ensuring they are informed whenever AI is being used in their diagnosis or treatment plan.
Widespread use in General Practice
The push for new legislation comes as AI is already being rapidly adopted across the UK. Current data suggests that approximately 40% of GPs are now using “AI scribes”—software designed to listen to and record patient consultations to save time on administrative tasks.
The recommendations move the UK toward a “lifecycle approach” to medical safety. Rather than testing a product once, the MHRA intends to gain new powers to monitor software throughout its entire time in use within the NHS, ensuring that the “learning” nature of AI does not introduce unforeseen risks to patient safety.
