Is AI Medical Transcription Safe and Compliant in the UK? What NHS England and the MHRA Expect in 2026
Manisha
Nov 18, 2026
Last reviewed: 18 November 2026 · Medically reviewed by Jeet Patel
A year ago, most clinicians asking about AI transcription wanted to know one thing: does it actually work? In 2026 the question has changed. Now it's "am I allowed to use it, and if something goes wrong, who carries the can?"
That shift is fair. Since January, NHS England has launched a supplier registry for AI scribes, published fresh guidance on using them safely, and the MHRA has set out when an AI scribe counts as a medical device. If you're a GP partner, a practice manager, a consultant running private clinics or a medico-legal expert, you'll have seen the headlines but probably not had time to read the documents behind them.
This guide does that for you. It explains what the rules say, how they apply to AI medical transcription as well as ambient scribes, and what to ask a supplier before you sign anything.
Yes, AI medical transcription can be used safely and lawfully in UK healthcare. You need four things in place: a data protection impact assessment, a supplier that meets UK data and clinical safety standards, clear information for patients (with a way to object), and a clinician who checks every output before it goes into the record.
AI transcription, AI scribes and ambient voice technology: what's the difference?
The terms get used interchangeably, but they describe different workflows, and the difference matters for risk.
| Type |
How it works |
Typical use |
| AI dictation transcription |
You dictate; AI turns your voice into a letter or report |
Clinic letters, radiology and cardiology reports, operation notes |
| Ambient AI scribe (AVT) |
The tool listens to the whole consultation and drafts notes |
GP and outpatient consultations |
| Hybrid AI + human transcription |
AI drafts; a trained medical transcriptionist checks it before it reaches you |
Specialist reports, medico-legal work, complex terminology |
NHS England and the MHRA use the phrase ambient voice technology (AVT) for tools that "capture and convert spoken words into text." Much of their guidance is written with ambient scribes in mind, but the principles data protection, clinical safety, human review apply just as much to dictation-based AI transcription. If you want a side-by-side on time savings, see our comparison of ambient AI scribes and traditional transcription.
What changed in 2026
Four developments shape the picture today:
1. January 2026 - the AVT supplier registry. NHS England published a self-certified registry of AI scribe suppliers, starting with 19 companies. Suppliers declare they meet standards on clinical safety, technology and data protection. NHS England only checks that submissions are complete; it is not an endorsement or a procurement framework.
2. April 2026 - NHS England guidance on safe use. Developed with input from the ICO and the National Data Guardian, it covers transparency with patients, objections, data retention and clinician responsibility.
3. 29 July 2026 - MHRA guidance on medical device status. It explains which AI scribe functions are regulated as medical devices and which are not (more on this below).
4. Government rollout plans. The government has said it will support AI scribe adoption in general practice, with framework procurement open to all NHS organisations in 2026–27, and has advised practices to appoint a clinical safety officer.
Adoption is already high. A survey of 598 UK GPs published in June 2026 found 43% were using an AI scribe. It also found that more than 60% were worried about inaccuracies and medicolegal risk, and that almost a quarter had tried one and stopped. In other words, the tools are popular, but trust is not settled.
Is AI transcription a medical device? What the MHRA says
This is the question that worried many practices most, and the MHRA's answer is reassuring for straightforward transcription.
Under the MHRA's guidance on AVT-enabled products, a tool is not a medical device if it only:
- transcribes speech into text
- summarises a consultation for the clinician to review
- formats information into structured lists for confirmation
- suggests clinical codes for terms that were explicitly said
- drafts letters or discharge summaries from information already in the record
It is likely to be a medical device if it suggests diagnoses or treatment, orders tests without clinician confirmation, or is marketed as something that "guides diagnosis." The MHRA is clear that a disclaimer won't change that: what the product does and how it's promoted decide its status. It also warns about feature creep. A tool that's fine today could become a regulated device after an update that adds "insights," so ask suppliers how they handle new features.
For most clinic letters, specialist reports and dictated notes, a transcription-only tool sits outside medical device regulation. That doesn't remove your other obligations, but it does simplify things.
Do patients need to give consent?
Not explicit consent, for direct care. NHS England's guidance says you don't need to ask patients to sign or formally agree before using an ambient scribe, because it is part of their individual care. What you do need is transparency:
- Tell the patient at the start of the consultation that an AI tool is helping with notes.
- Update your privacy notice, and consider posters, your website and leaflets.
- If the patient objects, don't use the tool for that consultation. The guidance is blunt: if someone dissents, "you cannot then use that tool."
The national data opt-out doesn't apply when the tool is used for individual care. Use judgement in sensitive consultations, where a patient may say yes out of politeness rather than comfort.
For dictation-based transcription, where you dictate a letter after the patient has left, there's no recording of the patient at all. That's one reason many consultants find dictation the simpler route.
UK GDPR and data protection: five questions to answer
The NHS England guidance says a DPIA is "highly likely to be a legal requirement" for ambient scribes, and it's good practice for any AI transcription service. Your DPIA should answer at least these:
1. Where is the data processed and stored? UK-based servers make the transfer question much simpler. Ask for the specific location, not just "secure cloud."
2. Who is the controller and who is the processor? Usually your organisation is the controller and the supplier is a processor acting on your instructions. Get it in the contract.
3. How long are audio and transcripts kept? NHS England expects recordings and transcripts to be deleted once the verified note is in the record, unless there's a justified reason to keep them.
4. Is patient data used to train the AI? It shouldn't be used for anything beyond the patient's care without a proper legal basis. Ask directly.
5. What security standards does the supplier meet? Look for the NHS Data Security and Protection Toolkit (DSPT), Cyber Essentials (ideally Plus), encryption in transit and at rest, and role-based access.
One small tip: if a UK supplier's website leans on HIPAA, that's a US law. It isn't a substitute for UK GDPR and DSPT evidence. We explore the technical side in more depth in protecting data in AI transcription.
Clinical safety: DCB0129, DCB0160 and your safety officer
Two NHS clinical risk management standards apply to health IT in England:
- DCB0129 is the supplier's responsibility. They should have a clinical safety case and hazard log, signed off by their own clinical safety officer.
- DCB0160 is yours. Your organisation assesses the risks of deploying the tool in your setting, keeps its own hazard log, and names a clinical safety officer.
For a GP practice this sounds heavier than it is. Ask the supplier for their DCB0129 documents, use them to build your DCB0160 assessment, and record how you'll monitor errors once you go live. Many ICBs and LMCs now offer templates.
Accuracy: the risk that matters most
Compliance paperwork protects you on paper. Accuracy protects your patients. Both NHS England and the MHRA say the same thing: the clinician remains responsible for every word that goes into the record. NHS England has also warned that practices may still be liable for negligence arising from AI use.
In our experience, AI transcription errors cluster in predictable places:
- Drug names and doses that sound alike (for example, hydroxyzine and hydralazine)
- Numbers and units such as milligrams versus micrograms
- Laterality left and right swapped in a fast dictation
- Negatives "no chest pain" becoming "chest pain"
- Speaker mix-ups in ambient recordings with relatives or interpreters
- Accents, background noise and poor connections
- Invented detail summaries that add something nobody said
None of this means AI transcription is unsafe. It means review has to be built into the workflow, not left to goodwill at the end of a long clinic. Practical steps that work:
- Read every output before signing it, with extra care for medications, numbers and sideNone of this means AI transcription is unsafe. It means review has to be built into the workflow, not left to goodwill at the end of a long clinic. Practical steps that work:s.
- Label AI-assisted entries where your system doesn't do it automatically.
- Audit a sample None of this means AI transcription is unsafe. It means review has to be built into the workflow, not left to goodwill at the end of a long clinic. Practical steps that work:of notes each month and log errors through your incident process.
- For high-stakes documenNone of this means AI transcription is unsafe. It means review has to be built into the workflow, not left to goodwill at the end of a long clinic. Practical steps that work:ts specialist reports, medico-legal reports, anything with dense terminology consider a hybrid service where a trained medical transcriptionist checks the AI draft before it reaches you.
The hybrid approach is the reason many consultants and reporting teams haven't gone "AI only." You keep the speed of AI but add a second pair of trained eyes, so your review becomes a final sign-off rather than a proofread.
Consultants, specialists and private practice: what's different
Most AI scribe guidance is written for GP surgeries. If you work elsewhere, a few points are worth knowing.
- Private practice: NHS England's guidance and the registry are aimed at NHS organisations, but UK GDPR, ICO expectations, your indemnity provider and GMC duties on clear, accurate records apply to everyone. Using the NHS framework as your benchmark is sensible, and it's what NHS-referring clients and hospitals will expect.
- Specialist reporting: radiology, cardiology and ophthalmology reports carry vocabulary that general AI models trip over. Ask suppliers how their models handle your specialty, and test with your own dictations.
- Medico-legal work: reports may be scrutinised in court, so accuracy and an audit trail matter more than turnaround. A human-checked medico-legal transcription service is often the safer choice.
- Scotland, Wales and Northern Ireland: NHS England's guidance covers England. The devolved nations have their own processes, though UK GDPR and MHRA rules apply UK-wide.
Supplier checklist: questions to ask before you sign
Print this, or send it to shortlisted suppliers:
- Is your product on NHS England's AVT self-certified supplier registry? (Relevant for ambient tools.)
- Can you share your DCB0129 clinical safety case and hazard log?
- Have you completed the DTAC, and do you hold a current DSPT and Cyber Essentials certificate?
- Where exactly is patient data processed and stored? Does it leave the UK?
- How long do you keep audio and transcripts, and can we set the retention period?
- Is our data used to train your models?
- Does any feature suggest diagnoses or treatment? If so, what is its MHRA status?
- How does output reach EMIS Web, SystmOne or our PMS? (NHS England now recommends EPR-integrated tools.)
- What accuracy monitoring do you provide, and is human review available for complex documents?
- How will you tell us about model updates or new features?
If you use EMIS or SystmOne, our guide to getting transcription straight into EMIS Web and SystmOne covers the practical set-up.
How iTranscript360 approaches AI transcription
We built our AI Transcription Services for UK clinicians who want speed without losing control of their records. Patient data is processed on UK-based servers under UK GDPR. You can choose fast AI-only transcription or our hybrid service, where experienced medical transcriptionists check the AI draft, with manual turnaround within six hours. Our team works across general practice, specialist reporting and medico-legal documentation.
If you're weighing up AI transcription for your practice or clinic, start a free trial with your own dictations, or request a quote for your workload.
Frequently asked questions
Is AI medical transcription legal in the UK? Yes. There's no ban. You need to meet UK GDPR, complete a DPIA, tell patients when an ambient tool is used, and make sure a clinician reviews every output. In England, NHS organisations should also follow NHS England's ambient scribing guidance.
Do I need patient consent to use an AI scribe? Not explicit written consent for direct care. You must tell patients the tool is being used and stop if they object.
Is an AI scribe a medical device in the UK? Not if it only transcribes, summarises, drafts letters or suggests codes for review. It becomes a medical device if it suggests diagnoses or treatment or acts without clinician confirmation.
Who is responsible if the AI makes a mistake? The clinician who signs off the note, and the organisation that deployed the tool. AI doesn't transfer clinical responsibility.
Is AI transcription as accurate as a human medical transcriptionist? For clear dictation in common specialties, modern AI is very good. For complex terminology, poor audio or medico-legal reports, a hybrid AI + human service gives an extra safety check.
What happens to the audio after transcription? NHS England expects audio and transcripts to be deleted once the verified note is saved, unless there's a justified reason to keep them. Agree retention with your supplier in writing.