Insights Ambient voice technology: the challenge facing providers, regulators and suppliers
AI scribes are moving from pilots to scale across the NHS. The benefits are real. The rules have just shifted. And the safety work now sits squarely with the people deploying them.
The sheer speed with which ambient voice technology has moved through the NHS is a strong indicator that there is a real and significant problem that needed solving. A clinician talks with a patient. The software listens, transcribes, and drafts the note, the letter or the codes. The clinician reviews it, edits it and signs it off. Clinicians ultimately spend less time typing, and more time directly with the patient. It’s both a convincing business case and a route to a more people-centric care model.
The evidence for the upside is strong. A GOSH-led evaluation across nine London sites and more than 17,000 encounters reported 23.5% more direct time with patients, appointments 8.2% shorter, and 13.4% more patients seen per A&E shift. The 10 Year Health Plan committed to a framework procurement, open to all NHS organisations, in 2026 and 2027. NHS England has told organisations to deploy AVT at pace, and regional deals now cover tens of thousands of clinicians.
But speed brings its own questions, and they land differently depending on where you sit.
When NHS England first published guidance in April 2025, AVT that used generative AI to summarise was treated as likely to be a medical device. In June 2025, NHS bodies were reportedly told to stop using AI transcription tools that did not meet national standards. A self-certified Supplier Registry followed in January 2026.
Then, on 29 July 2026, the MHRA clarified its position. AVT intended solely for transcription, summarising conversations, drafting letters or suggesting codes for a clinician to review is not a medical device. AVT that supports diagnosis or treatment, or takes action automatically, is. The MHRA was clear that this does not change the law, and that clinicians remain responsible for reviewing what the software produces.
On the same day, NHS England set out what boards should assure: oversight and accountability, local governance across clinical safety, information governance and digital assurance, staff training, risk management including automation bias and incident reporting, and contracts that cover performance monitoring, audits, functionality changes and model updates.
For most AI scribes, the safety case now lives with the organisation using them.
For many scribes, there is now no device regulation doing the heavy lifting. That makes DCB 0160 and local governance the main line of defence. Trusts, ICBs and practices have to understand the hazards (a hallucinated medication, a missed allergy, a note attributed to the wrong patient), decide which controls they rely on from the supplier and which they own, train staff to review outputs properly, and monitor what happens in use. Many are doing this for several products at once, with stretched clinical safety capacity.
The questions to ask are practical. Does the supplier’s hazard log make clear which controls we own? What happens when the supplier updates its model? How will we know if accuracy drifts? How do we report an incident, and to whom?
Regulators are balancing two real risks. Over-regulate, and a technology with clear benefits for patients and staff stalls. Under-regulate, and harm goes unseen, because routes for reporting AI-related incidents are still immature. HSSIB opened a national investigation into AVT in acute hospitals in August 2026, noting that the safety implications are not fully understood. The King’s Fund has pointed to a lack of transparent performance metrics and called for urgent work on how AVT safety reporting interacts with the MHRA Yellow Card scheme. And products evolve: a scribe that only drafts notes today may suggest diagnoses tomorrow, crossing into device territory with a single release.
For suppliers, the bar has moved rather than dropped. Buyers still expect DCB 0129, a clear clinical safety case, DTAC, DSPT and Cyber Essentials. They will increasingly expect evidence of accuracy in use, a clear account of which models you depend on, and a change process they can trust. The National Commission recommended that contracts explicitly allocate responsibility for every required risk control between supplier and provider. The suppliers that win at scale will be the ones that make the provider’s safety work easier.
AVT is a clear case of the problem Qyro was built to solve. One product, many providers, each needing to understand and own its share of the risk. In Qyro Hazard Studio, a supplier’s hazards carry across to each provider as data, with inherited and local controls kept apart. Qyro Assurance Planner turns post-deployment monitoring into dated, owned steps. And when a supplier changes its product, both sides can see what it touches.
Done well, AVT gives clinicians time back and makes care safer. Getting there depends on suppliers, providers and regulators each doing their part, and on the handoffs between them holding.
