MHRA Annual Report Signals Regulatory Turning Point for AI in UK Healthcare

The Medicines and Healthcare products Regulatory Agency published its 2025–26 Annual Report on 15 July, and the document reads less like a routine accountability exercise and more like a strategic declaration. The headline numbers are impressive: record clinical investigation approvals, a 17% rise in device trial sign-offs. But it is the AI chapter that should command the attention of anyone building or buying health technology in the UK right now.

From sandbox to system

The report confirms that the MHRA's AI Airlock, the world's first regulatory sandbox for AI as a medical device, has completed its second phase and secured renewed multi-year funding, with Phase 3 now in development. The Department of Health and Social Care committed £1.2 million per year from 2026 to 2029 to sustain the programme. Across its pilot and second phase, the Airlock explored some of the most technically demanding territory in the field: large language models in clinical decision support, synthetic data for radiology validation, and real-time post-market surveillance monitoring.

The MHRA has not stopped there. In June, the regulator unveiled two further sandboxes within two days. The first, announced at London Tech Week on 9 June by Science Minister Lord Vallance, is a UK-wide medicines sandbox giving companies and researchers a controlled environment to test AI tools for predicting drug safety. It is a direct response to the fact that adverse drug reactions send around 250,000 people to hospital annually and cost the NHS over £2 billion a year. The second, London Region I, is a live-deployment programme run in partnership with NHS England and London's Health Innovation Networks. Up to 10 AI medical device manufacturers will be selected for its initial phase, working directly alongside NHS providers under MHRA oversight. Expressions of interest were invited from manufacturers this month.

The Commission's summer deadline

Running parallel to this sandbox activity is the National Commission into the Regulation of AI in Healthcare, established in September 2025 and chaired by Professor Alastair Denniston. Its recommendations are expected before the end of summer 2026. The MHRA published two substantial evidence reports in June: one synthesising a public call for evidence that attracted 760 individual and institutional submissions, the other drawing on wider research and engagement with patients, frontline staff, and industry partners.

The overarching message from that evidence base: broad recognition of AI's potential benefits, paired with a firm expectation of ongoing monitoring, transparency, accountability, and human oversight. Notably, 89% of non-using clinicians in one study cited professional liability and medico-legal risk as a major concern, and those worries were not confined to non-adopters. Even experienced users shared them. That finding has direct implications for whatever framework the Commission recommends. Liability allocation, in particular, remains the thorniest unresolved question.

The MHRA has also formalised Predetermined Change Control Plans (PCCPs) for AI and software medical devices as part of its draft Medical Devices (Amendment) Regulations 2026, published in May. PCCPs allow manufacturers to define anticipated software modifications at the point of initial regulatory submission: a practical concession to the reality that adaptive AI systems cannot be governed by static approval processes.

What the NHS is deploying right now

While the regulatory architecture is being assembled, the NHS has not waited. In early July, NHS England set out how £10 billion of technology investment, allocated by government last year, will be deployed over the next three years. The centrepiece is an AI triage tool in the NHS App, due to reach more than 200,000 patients in the next 12 months and all App users by April 2028. A trial at a GP practice in Sussex cut phone queue volumes by 29%. Ambient voice technology, which records consultations and generates real-time clinical summaries, is being rolled out across outpatient departments at multiple NHS trusts. A GOSH-led study found the technology freed clinicians to spend nearly a quarter more time with patients, and modelled that a national A&E rollout could create space for over 9,000 additional consultations every day.

NHS deployment is accelerating precisely as the Commission prepares the rulebook that should govern it. That sequencing is not unusual in healthtech; regulators have rarely moved faster than the technology they oversee. But it concentrates the stakes considerably on whatever the Commission publishes.

A gap the market has yet to close

For commercial players, the proprietary data tells its own story. AI Business Dispatch analysis of Companies House and IPO data (as of July 2026) shows only 53 new hospital-activities companies (SIC 86.10) were incorporated in Q3 2026, down 83% on the prior period. Class 44 UK trademark filings, covering medical and health services, fell 80.1% in the same period, to just 457. Most striking: 98.3% of active SIC 86.10 companies hold no Class 44 trademark whatsoever. In a sector where brand trust is inseparable from clinical trust, that figure suggests the majority of providers have not yet begun to think commercially about their AI-facing identity.

The MHRA has been explicit that regulation should be an enabler, not a barrier. CEO Lawrence Tallon told Parliament the past year had been spent "maintaining performance and advancing capabilities to be a proportionate and agile regulator." The AI Commission's recommendations, when they arrive, will test whether that ambition produces a framework the other 98.3% can actually use.