UK regulator's AI scribe rule turns on marketing, not medicine

The UK's medical device regulator drew a line this week on a category of clinical AI already in use across the NHS, and the line runs through how the product is sold, not what it does. AI scribes — note-taking software that listens in on a patient visit and drafts the doctor's clinical record, referral letter, or billing code — sit at the centre of the question, the STAT+ Health Tech newsletter reported.

On 29 July 2026 the Medicines and Healthcare products Regulatory Agency published new Ambient Voice Technology (AVT) guidance, developed with NHS England. Products that only transcribe, summarise clinical conversations, draft letters, or suggest codes for a clinician to review are not regulated as medical devices under the current framework, and that will not change, the agency said. Products intended to support diagnosis, treatment, or prevention, or that take automated action such as placing orders without clinician review, are medical devices and must meet the relevant safety and performance requirements.

The split turns on the manufacturer's stated "intended purpose" in instructions for use, labelling, and promotional materials. Disclaimers are not effective if other claims contradict them, the regulator noted. A vendor that markets the same product as "ambient documentation" gets one regime. The same vendor describing it as "clinical decision support" gets a different one. The boundary is drawn at the marketing, not the mechanism.

NHS England rewrote its own AVT guidance in the same week to remove its prior explanation of how the UK Medical Devices Regulations 2002 applied to AI scribes. Until now, NHS England treated all AVT products as medical devices, and suppliers needed MHRA Class I registration to appear on the NHS-approved supplier registry. The agency framing is now the national one, and the registry expectation has dropped with it. Specialist medical-device law firm Bristows flagged the shift as a marked departure from the prior framework.

The Royal College of Physicians broke with the regulator publicly. The college's clinical digital health lead, Dr Anne Kinderlerer, warned that under-regulating AVTs could weaken oversight of clinical decision support and shift liability onto individual clinicians and trusts, the college said in its statement. Pulse Today reported doctor leaders warning of greater risk to GPs as a result. A tool that drafts the note a doctor reads before deciding effectively shapes the diagnosis, in the RCP's view, whatever the vendor calls it. The "transcription versus decision support" line holds on paper and collapses at the bedside.

MHRA CEO Lawrence Tallon framed the guidance as removing ambiguity so lower-risk tools can spread faster while protections stay where they matter. NHS England Chief Digital and Data Officer Rob Thompson said the technology will now be rolled out across the NHS "safely, consistently and at scale." The clinician's responsibility to review and verify every AI-generated transcript, summary, and code suggestion before it touches patient care is unchanged.

Two structural limits sit underneath the announcement. The guidance is interpretive, not new law, so the boundary will be tested in the next contested classification call. And the regime stops at the Irish Sea: Northern Ireland remains under EU MDR 2017/745, where the EU's stance on ambient voice tools is still unresolved, so a single vendor with a Great Britain and Northern Ireland footprint now sits under two rule books in one country.

MHRA, NHS England, and the National Commission into the Regulation of AI in Healthcare are working on a future framework. The next move worth watching is not another guidance note. It is the first contested intended-purpose case.