UK watchdog calls for new AI regulations in NHS as technology becomes routine

Britain's medicines regulator has concluded that the country needs an entirely new set of rules for artificial intelligence in healthcare. The Medicines and Healthcare Products Regulatory Agency, which oversees all medical devices and drug approvals in the UK, published 44 recommendations this week to overhaul how it governs AI products as they move from experimental tools into routine NHS practice. Lawrence Tallon, the MHRA's chief, told the BBC that patients will increasingly encounter AI as a normal part of their care—but only if regulators can build frameworks that preserve public trust in the process.

The recommendations emerged from an independent commission that gathered input from more than 12,000 people: patients, clinicians, researchers, and others working inside the health system. The core proposals are straightforward in principle but demanding in execution. Patients should have the right to know when AI is involved in their care and to access clear information about which products are being used. Regulators should be able to continuously monitor AI systems after they've been approved and remove them from use if they malfunction or lose effectiveness over time. Developers who fail to meet required standards should face penalties. And there should be an "L plate" trial system—borrowing the learner-driver metaphor—that would let new AI models be tested by healthcare professionals under close supervision before wider deployment.

Tallon emphasized that current medical device regulations are fundamentally mismatched to AI's nature. Those rules were built for things like hip replacements and knee replacements, for stethoscopes and plasters—products that remain stable after they leave the factory. AI systems are different. They continue to change after approval. New data flows in. They learn. They adapt. They drift. A model trained to spot tumors on scans might perform well on the training data but behave unpredictably when exposed to images from a different hospital, a different population, a different machine. Regulators have no precedent for this kind of moving target.

One immediate flashpoint is AI scribes—software powered by large language models that record GP consultations and generate clinical notes. About 40 percent of UK-based general practitioners now use them. But recent research from the University of Edinburgh found that patients are less willing to share sensitive information—details about substance abuse, mental health struggles, sexual history—if they know an AI system is processing the conversation. Professor Henrietta Hughes, a GP who contributed to the regulatory report, said many of her patients accept AI in their consultations, but some explicitly opt out. "Some say, 'I don't want to talk to a robot', and that is also fine," she said. She acknowledged that the scribes sometimes make errors in their notes but argued it remains the doctor's responsibility to catch and correct them.

The stakes of getting regulation right are high. Some researchers see AI as transformative for medicine. Professor Alastair Denniston, an ophthalmologist who worked on the report, called it "an exceptional opportunity" for healthcare, potentially as significant as the introduction of antibiotics or MRI scanning. Industry leaders are bullish: Rene Haas, the chief executive of chip designer Arm, told the BBC in September that he believes AI will produce a cure for cancer within this generation. But those possibilities sit alongside real risks. AI models can perpetuate or amplify biases present in their training data, leading to worse outcomes for certain patient groups. AI chatbots have been known to offer incorrect medical advice with unwarranted confidence.

Tallon acknowledged that no country has yet solved the problem of AI regulation comprehensively. This is a global challenge, and the frameworks being built now will shape how the technology is deployed across healthcare systems for years to come. The MHRA's 44 recommendations are a starting point—a signal that Britain recognizes the gap between the regulatory tools it has and the tools it needs. Whether Parliament acts on them, and how quickly, will determine whether AI becomes a trusted part of NHS care or a source of public anxiety.