MHRA's AI Rulebook Is Almost Here - and the NHS's £10bn Voice-Tech Bet Hangs on It

A Regulator Playing Catch-Up at Speed

Something unusual is happening in UK healthtech regulation. The rules are being written after the contracts.

A GP surgery in Hertfordshire signed a three-year contract for AI-drafted consultation notes before regulators had resolved their classification differences, and in April 2025 NHS England and the MHRA tightened the rules for ambient voice technology, requiring any tool that goes beyond transcription into summarisation to register as Software as a Medical Device. Within months, the MHRA had established a national commission specifically tasked with rewriting those very rules.

That commission, the National Commission into the Regulation of AI in Healthcare, is now past its stated deadline. It was tasked with publishing recommendations by summer 2026. Chaired by Professor Alastair Denniston, with the Patient Safety Commissioner as deputy, it received more than 770 responses to its call for evidence before closing in February. The sector is waiting.

The NHS Has Not Waited

On 4 July, NHS England announced a major acceleration of AI tools across the health service, with a new AI triage tool in the NHS App and widespread access to AI notetaking tools among the improvements being prioritised. The announcement set out how £10 billion of funding over the next three years would be used for a major overhaul of the health service's technology, digital and data systems.

The App's AI triage tool is due to reach more than 200,000 patients within 12 months and be available to all NHS App users by April 2028; an initial trial at a GP practice in Sussex produced a 29% reduction in the number of people queuing on the phone.

The notetaking piece is arguably the more consequential. There will also be an England-wide rollout of AI tools that record conversations between patients and NHS staff to generate real-time transcriptions and clinical summaries. More than half a million NHS staff are also being given access to AI tools that could free up an average of two days every month from admin duties, with NHS England providing 505,000 clinicians and support staff with access to Microsoft 365 Copilot.

Ambient voice technology operating at clinical-summary level sits squarely inside the MHRA's contested classification territory.

The Regulatory Architecture, Still Under Construction

On 11 May 2026, the MHRA released the draft Medical Devices (Amendment) Regulations 2026, which introduce the International Reliance Pathway, formalise Predetermined Change Control Plans for AI and software devices, strengthen post-market surveillance obligations, and align IVD classification rules with globally recognised risk-based frameworks.

The PCCP mechanism matters directly to AI software vendors. A manufacturer can set out potential prospective changes it envisages making to a Software as a Medical Device in a Predetermined Change Control Plan, and provided those changes stay within that plan, the manufacturer may implement them without triggering a full re-certification by a UK approved body. For large language model-based scribes, which iterate constantly, this is the difference between viable commercial deployment and regulatory paralysis.

The draft amendment is expected to be adopted in December 2026 and come into force in June 2027, subject to parliamentary approval. Until then, suppliers operate under a patchwork: all ambient voice tools that generate summarisation must hold at least MHRA Class 1 medical device status, while solutions producing generative diagnoses or management plans require at least Class IIa approval.

As of late June 2026, TORTUS became the first ambient voice technology certified to UKCA Class IIa, assessed by approved body Scarlet. Most of its competitors remain self-declared Class I.

What the Market Data Shows

The proprietary data tells a sobering story about sector readiness. According to AI Business Dispatch analysis of Companies House and IPO data (as of July 2026), just 71 new SIC 86.10 hospital-activity companies were incorporated in Q3 2026, a drop of 77.2% against the prior period. Class 10 UK trademark filings, covering medical devices and instruments, fell 71.3% over the same period, to 359 filings. Most striking: 99.4% of active SIC 86.10 companies hold no Class 10 trademark whatsoever.

That figure is not simply a legal curiosity. A company deploying AI tools that touch clinical documentation or patient triage without formal device classification or brand protection is operating with thin governance infrastructure at exactly the moment regulators are tightening their scrutiny. The trademark gap may also signal a market dominated by software-as-a-service entrants who have categorised their products as administrative tools rather than medical devices, a distinction the MHRA is actively narrowing.

The Safety Question Regulators Cannot Escape

The Commission's call for evidence surfaced serious concerns about safety, oversight and liability: who bears legal responsibility when an AI-drafted note contains a material error, how algorithmic performance across regional accents and dialects is monitored, and at what point a tool classified as administrative assistance begins, in practice, to shape clinical decision-making.

On 11 June 2026, the MHRA published two reports describing the findings of an extensive engagement and evidence-gathering programme to inform future rules on how AI can be used safely in healthcare. Those reports feed directly into the Commission's still-unpublished recommendations.

In the care sector, adoption is already ahead of governance. A survey of 122 homecare providers found that 70% are already using AI in some form, with that figure projected to reach 85% within a year. Three-quarters said AI had improved care quality, and among those re-inspected since adopting it, 59% saw their CQC rating improve.

The MHRA has built something genuinely novel in the AI Airlock. Across its pilot and second phase, the programme has explored large language models in clinical decision support, synthetic data for radiology validation, and real-time post-market surveillance monitoring. The Department of Health and Social Care will provide £1.2m annually from 2026 to 2029 to support broader evaluations and advance sustainable regulatory pathways for AI healthcare technologies.

The Commission's recommendations, when they arrive, will land into a market that has already placed its bets. Whether the rulebook vindicates them is the question the sector cannot yet answer.