MHRA Clarifies Rules for NHS Ambient Voice Tools
On 29 July 2026, the Medicines and Healthcare products Regulatory Agency, working with NHS England, published new guidance on ambient voice technology in health and care settings in Great Britain. If that sounds technical, the key point is simple: the regulator is trying to spell out when an AI note-taking tool is just an admin aid, and when it starts to count as something more serious in law. That matters because the NHS is using more of these products, often as part of a wider push to bring more AI into everyday care. The guidance is meant to give developers, suppliers and NHS organisations clearer rules before these tools spread further.
Ambient voice technology, often shortened to AVT and sometimes called an AI scribe, listens to a conversation between a clinician and a patient and turns it into notes, summaries or draft letters. Some products can also suggest clinical codes for a clinician to check. You can see why busy services are interested: less time typing could mean more time with patients. For readers trying to picture it, think of AVT as a digital assistant for paperwork rather than a digital doctor. It may help capture what was said in a consultation, but that does not automatically mean it should make clinical decisions.
The MHRA says that, under the current framework, an AVT product is not regulated as a medical device if it is only intended to transcribe conversations, summarise them, draft letters, or suggest codes for a clinician to review. In other words, if the product stays on the documentation side of care, it does not fall under medical device rules simply because AI is involved. That does not mean anything goes. The regulator says NHS organisations still need to think carefully about safe deployment, and NHS England has issued separate guidance for health and care settings in England on using AI-enabled ambient scribing products.
The position changes when the tool starts to shape clinical care or act on its own. If an AVT product is intended to support diagnosis, treatment or prevention, or if it takes automated action such as placing orders without clinician review, the MHRA says it is regulated as a medical device and must meet the relevant safety and performance requirements. That is one of the most important lines in the whole announcement. A system that helps write up a consultation is being treated differently from a system that influences care or triggers action. The guidance does not create a new law; it explains how existing medical device law applies to this fast-moving technology.
**What this means:** clinician oversight is still the part you cannot skip. The guidance says doctors and other clinicians remain responsible for reviewing and verifying AI-generated transcripts, summaries and other outputs before they are used in patient care. That responsibility has not changed. For patients, that should be reassuring. For staff, it is also a reminder that an AI note is not finished work. If a summary is wrong, misses detail or adds something that was never said, a human clinician still has to catch it before it reaches the patient record or affects care.
The MHRA and NHS England are also putting real weight on local governance. NHS boards and executive teams are being asked to make sure any AVT deployment comes with proper clinical oversight, staff training, procurement checks and clear internal processes. They are also expected to notice when a product's functions change, because a tool that starts as a scribe could move into regulated territory if it begins making or carrying out clinical decisions. This matters more than it might first appear. In AI, software can change quickly. A product update may seem minor, but it can shift a tool from admin support into something with much bigger safety consequences.
In public statements, MHRA chief executive Lawrence Tallon said the aim is to remove ambiguity and give NHS organisations, suppliers and clinicians clearer ground rules. Rob Thompson, NHS England's chief digital, data and technology officer, said the service now plans to roll out the technology safely, consistently and at scale. Supportive responses also came from Haris Shuaib of Newton's Tree, Professor Mary Dixon-Woods of THIS Institute at the University of Cambridge, Mark Cubbon of Manchester University NHS Foundation Trust and Adam Igra of Cambridge University Hospitals NHS Foundation Trust. Taken together, those comments show broad support for the direction of travel. But the praise comes with an obvious condition: the faster the NHS adopts AI tools, the more careful it has to be about where admin help ends and regulated clinical support begins.
The MHRA says the guidance was developed with input from NHS clinicians, technologists and patient safety advocates, and that it reflects the direction of the National Commission into the Regulation of AI in Healthcare. That tells us this is not the final word on AI in health. It is an attempt to give the system practical rules now, while a bigger future framework is still being worked out. If you are watching AI in the NHS, that is the real takeaway. The question is no longer whether these tools will be used, because many already are. The harder question is who checks them, how they are classified, and whether patient safety stays ahead of marketing claims.