Summary
The 10 Year Health Plan for England sets an ambition to “make the NHS the most AI-enabled care system in the world”. In line with this aim, there has been a significant increase in use of ambient voice technology (AVT) products, sometimes referred to as AI scribes, by healthcare providers.
Responding to these developments, the Health Services Safety Investigations Body (HSSIB) has recently launched a new investigation exploring the use of AVT in hospitals. In this blog, Patient Safety Learning considers:
Benefits and risks associated with the use of AVT products in healthcare.
The evolving regulatory environment around this technology in the NHS
Key issues for a new patient safety investigation to consider.
Content
When looking at the use of AI products in healthcare, the application of AVT is a significant area of growth and interest.
AVT products are AI-enabled systems that capture and record natural conversations in the background. When combined with generative AI, they can convert speech directly into structured documentation, such as notes and letters.
Benefits and risks
There are various benefits that can be gained by using AVT products in healthcare, such as:
Providing greater accuracy in notetaking.
Reducing the time healthcare professionals spend on administrative tasks.
Making administrative processes more effective through integrating the data they collect with electronic patient record systems.
Improving conversations between patients and healthcare professionals, removing the need to focus on notetaking/looking at a screen during consultations.
A recent study tested across nine NHS sites demonstrated some of these benefits in practice.[1] Looking at the implementation of a specific AI-scribe tool, it found a 23.5% increase in direct patient interaction time during appointments. Alongside this, the results showed an 8.2% reduction in overall appointment length when AI-scribes were used.
However, balanced against these benefits, the growing use of AVT also introduces new patient safety risks, such as:
AI hallucinations—where an AVT product creates false or misleading information presented as a plausible statement.
Transcription errors—potentially resulting in incorrect subsequent patient care and treatment if not identified.
Data omissions—key information being left out of care records.
Evolving regulatory environment
Initially it had appeared that AVT products in healthcare would be monitored and regulated as medical devices under the remit of the Medicines and Healthcare products Regulatory Agency (MHRA). Early guidance from NHS England indicated these should be considered as Class I medical devices.[2] [3] Class I is the lowest of four classes, with such devices “generally regarded as low risk”.[4]
However, there has been debate about whether this classification is appropriate. Particularly as the AI summarisation of an AVT transcript could potentially change the meaning of what was said, with potentially serious implications for future care and treatment. Some people, therefore, had assumed that it would be classified as a Class IIa medical device, “generally regarded as lower medium risk”. Indeed, one of the first AVT companies in the UK recently achieved that certification level for its product.
Subsequently, new guidance published by the MHRA at the end of July 2026 clarified that only AVTs that have an “intended purpose” that meets medical device certification requirements should be regulated as Class I devices.[5] By “intended purpose” the MHRA means this is what the manufacturer says the AVT is designed and intended to be used for (e.g. to transcribe speech, support clinical diagnosis, etc).
Some AVT tools just do the administrative tasks. They write down what was said, shorten it, draft letters or suggest the right codes to be used. The healthcare professional still reviews this information and makes a decision on what it means. Tools like this are not doing anything medical, so the MHRA medical device rules will not apply to them.
But, if a tool is doing something medical it counts as a medical device, and then it has to be risk-assessed. Most of these would be Class I devices. The exception is a tool that actually works out what is wrong with the patient, or gives the doctor the key piece of information they need to make that decision. Then it isa Class IIa device, which has to pass much more rigorous testing by the MHRA.
HSSIB announcement
Shortly after the publication of the MHRA’s new guidance, HSSIB announced it would be launching a new investigation exploring the use of AVT in hospitals.[6] HSSIB investigates patient safety concerns across the NHS in England and in independent healthcare settings where safety learning could help to improve NHS care.
Focusing on acute adult secondary care settings, their investigation will seek to understand:
How AVT has contributed, or may contribute, to patient harm.
The patient safety risks associated with AVT.
How accountability for AVT is understood locally and nationally.
Their subsequent investigation report is expected to be published in summer 2027.
Key patient safety issues to explore
At Patient Safety Learning we welcome proposals by HSSIB to look at the patient safety issues around the implementation of AVT products in the NHS.
We are supportive of the safe use of AVT in healthcare. We also acknowledge the workload and burnout issues that staff are currently facing, and that early AVT evaluations point to genuine gains in direct patient contact time. However, AVT products are also being implemented at pace in healthcare organisations in the absence of a significant evidence base on how they work in settings other than primary care.
Below we highlight some key issues we believe that HSSIB should consider as part of their forthcoming patient safety investigation.
Differences between care settings
Currently much of the evidence around the application of AVT products in healthcare has been focused on primary care and outpatient care. Therefore, we welcome HSSIB looking more broadly at acute secondary care settings.
In particular, we think there would be value in exploring the difference between AVT products in quieter, one-on-one settings such as consultations, and in busier environments, such as Accident and Emergency departments. The latter are likely to have greater background noise and more interruptions, potentially impacting the performance of an AVT and the likelihood of errors occurring.
Errors resulting in safety risks
As mentioned earlier, AI hallucinations, transcription errors and data omissions by AVTs all carry with them potential risks to patient safety if not identified. We believe it would be useful for HSSIB’s investigation to separate out these types of errors, considering their frequency and impact on safety.
We would also suggest significant consideration is given to the potential for look-alike/sound-alike errors, where a medication or treatment is incorrectly recorded as a similar sounding alternative. We already know about the harm such errors can create when combined with electronic systems, as a recent National Patient Safety Alert about the incorrect recording of penicillin allergy as a penicillamine allergy highlighted.[7]
Role of the healthcare professional
When considering safety risks with AVTs, it is important to consider how they interact with the staff who are using these tools.
A significant issue HSSIB should consider is automation bias: the human habit of trusting computer or AI decisions too much. This is particularly important as existing NHS England and MHRA guidance around AVTs appears to rest on the assumption it is supporting, not replacing, clinical judgement with the clinician responsible for checking output.
There is some emerging evidence that suggests that staff (particularly more junior staff) can often quickly defer to AI systems. For example, a pilot study comparing radiology teams in Scotland and England found that less experienced staff under time pressure increasingly deferred to AI outputs.[8] Participants also noted that no UK medical school currently includes compulsory AI or data-science training, limiting clinicians’ ability to act as informed overseers rather than being passive end-users.
This connects to the broader issue of needing to ensure staff have the sufficient digital skills and training to utilise AVT products and their outputs safely.[9]
Role of the patient
In too many discussions around the implementation of AVT in the NHS, the role of the patient is largely absent. Patients and families need to be confident they can trust healthcare providers to use AI safely. We are now beginning to see cases emerge of complaints linked to patients not being told AI tools are being used in their case.[10]
As part of their investigation, a key question we think HSSIB should consider is how consent is obtained, recorded and honoured in acute settings, and what happens operationally when a patient refuses? Recent research by Healthwatch England suggests that patients are more comfortable talking about generalist topics with an AI scribe in the room, but less so on sensitive issues such as sexual health, mental health and domestic violence.[11]
HSSIB may wish to consider as part of this whether AVT products, once in place, have clear options for patients to make viable choices on their use and opt out as needed.[12]
Biases
As part of this investigation, we believe it would also be useful for HSSIB to explore whether there may be any safety issues associated with accent-related errors.[13] [14] This could be particularly an area of concern where patients do not speak English as a first language, where communication is in other languages or where patients have strong accents. In increasing the rollout of AVT products in healthcare services, we need to ensure they are not reinforcing existing or introducing new biases that increase patient safety risks for specific groups.
Regulation
Earlier in this article, we set out how assumptions on how AVT products are to be regulated have evolved over the past year.
NHS England has established an AVT Supplier Registry, intended to help organisations with the procurement and assurance of these tools.[15] This is a self-certified system, meaning a supplier must declare that its product meets the requirements for inclusion on the NHS England registry. Currently there is no independent verification mandated by NHS England or any other body. The HSSIB investigation may wish to consider the implications this has for ensuring the accuracy, performance and safety of AVT products before they come into use.
AVT products also need to conform with existing clinical assurance standards and processes in the NHS: DCB0129 and DCB160.[16] This will mean that every deploying trust needs a DCB0160 clinical safety case for AVT, informed by the supplier's DCB0129 work. Given the likely speed and scale of regional rollouts, we believe HSSIB should:
Test whether these clinical safety cases have been developed.
Consider whether Clinical Safety Officers have the capacity and AI expertise to produce them.
Look at whether hazard logs associated with these cases include ways in which an AVT can fail in clinical use.
Detecting, reporting and acting on harm
Currently there is no reliable mechanism to detect, report or count AVT-related patient safety incidents.
For the AVTs not being classed as a medical device, they would not be reportable through the MHRA Yellow Card Scheme.[17] Incidents could be reported through the Learn from patient safety events (LFPSE) service, the national NHS system for recording and analysing patient safety events in healthcare.[18] However LFPSE currently has no established route for recording that an incident involved AI-generated documentation.
This is not just an issue for HSSIB to consider, but something NHS England should be actively seeking to address now. We consider that we are currently committing to significant rollouts of AVT software without having in place adequate mechanisms for capturing, analysing and acting on any potential avoidable harm associated with these in use.
NHS England in the Midlands recently announced it had procured at scale an AVT solution for all of its 1,239 GP practices as well as over 70,000 clinicians across 15 acute and community trusts.[19] There is currently no national requirement to demonstrate safety outcomes before scaling to this level.
Sharing learning
This article has focused on HSSIB’s investigation, which is limited by their statutory remit to England. However, the patient safety issues highlighted here would also be applicable to the implementation of AVT products in Scottish and Welsh healthcare settings. While there are differences in regulatory arrangements in Northern Ireland (which follows the European Union’s Medical Device Regulation), many of the broader issues raised here are equally pertinent.
We would anticipate that the findings of HSSIB’s investigation to have important insights for patients, healthcare professionals and organisations across the UK. We would, therefore, expect to see the Department of Health and Social Care and MHRA subsequently working with key stakeholders in the devolved administrations to share insights and recommendations from this investigation as appropriate.
Concluding thoughts
Patient Safety Learning recognises the potential of AVTs to improve patient care and productivity in the NHS. However, we believe it is vital that patient safety considerations are hardwired into the procurement and implementation of these solutions.
This is not an exhaustive list of issues for HSSIB to consider within its investigation, but we hope it is a helpful way of shaping discussion on the remit of this review. We would welcome the opportunity to contribute to this directly in due course.
References
NHS England procurement guidance stated suppliers must hold Class I accreditation for AVTs. NHS England. Supplier AVT Self Certified Registry for the use of AI-Enabled Ambient Scribing Products in Health and Care Settings. 29 October 2025.
MHRA. Borderline products: classifying medical devices and risk. 11 July 2023.
A recent report from The King’s Fund identifies staff capability and capacity as one of the key five enablers that determine whether AVT incentives and contracts can be successfully translated into impact. The King’s Fund. Seizing a rare opportunity: scaling ambient voice technology in the NHS. 23 July 2026.
Health Service Journal. First complaints made over clinician use of AI. 6 July 2026.
Healthwatch England. What does the public think about AI scribe use in healthcare. 16 July 2026.
NHS England. Ambient Voice Technology Self-Certified Supplier Registry. Last accessed 13 August 2026.
These standards provide essential requirements for manufacturers of health IT systems and healthcare providers in assessing and managing clinical risks to ensure the safety of digital solutions in the NHS. NHS England. Review of digital clinical safety standards: DSC0129 and DCB0160. Last accessed 2 February 2026.
Through the Yellow Card Scheme the MHRA collects and monitors information on suspect concerns involving healthcare products and adverse events involving medical devices. MHRA. Yellow Card Scheme. Last accessed 13 August 2026.
NHS England. Midlands leads the way on ambient voice technology. 15 July 2026.
Recommended Comments
Create an account or sign in to comment