MHRA's London AI Sandbox Goes Live as Care Homes Struggle to Bridge the Adoption Gap

The gap between regulatory approval and real-world NHS deployment has long been the graveyard of promising healthtech. This summer, the MHRA is trying something different.

The London Sandbox: Regulation as Enabler

In late June, the MHRA launched what it calls the London Region I programme: a regulatory sandbox letting up to ten AI medical device companies deploy their tools in live London NHS settings under direct regulator oversight. The programme runs until December 2028 and was developed with NHS England (London) and the capital's three Health Innovation Networks: Imperial College Health Partners, UCLPartners and HIN South London.

The logic is simple but the execution is not. Rather than approving devices in the abstract, the sandbox generates the kind of procurement-grade real-world evidence that NHS trusts actually require before they'll write a purchase order. As MHRA chief executive Lawrence Tallon put it, the initiative aims to demonstrate "that regulation can be an enabler for innovation, not a barrier."

The MHRA will invite expressions of interest from NHS bodies and manufacturers imminently. Critics will note the obvious constraints: ten companies, one region, a two-year timeline. Whether lessons from a London pilot can translate into national adoption frameworks is the real test, and it's one the NHS has failed before.

This sits alongside a parallel MHRA sandbox announced at London Tech Week in June, focused specifically on medicines safety. That initiative will explore how AI can better predict adverse drug reactions, a problem that sends around 250,000 people to hospital annually in the UK and costs the NHS over £2 billion a year. Up to five AI-driven approaches will be tested in the first phase, with industry and academic partners being brought in from summer 2026.

A Framework Still Taking Shape

Both sandboxes feed into a broader regulatory overhaul that is, frankly, overdue. The current UK framework does not specifically address AI, leaving developers working within general medical device rules never designed for adaptive, continuously learning systems.

The MHRA has confirmed it will publish a dedicated regulatory framework for AI as a medical device in 2026, as set out in the government's Life Sciences Sector Plan. That framework is expected to introduce structured requirements for AI lifecycle governance, transparency and cybersecurity. Separately, an International Reliance Framework due by Autumn 2026 will allow manufacturers holding approvals from the US FDA, Health Canada or Australia's TGA to use those as the basis for a streamlined UK application, potentially cutting time to market by six to twelve months.

Meanwhile, the National Commission into the Regulation of AI in Healthcare, chaired by Professor Alastair Denniston with England Patient Safety Commissioner Professor Henrietta Hughes as deputy, completed its call for evidence earlier this year. Its recommendations to the MHRA are expected in 2026 and will directly shape the forthcoming AI-specific framework.

On the operational side, the MHRA has already waived fees for micro and small UK firms in a clinical investigation pilot running since January 2026. Average approval times for clinical investigations in 2025 hit 51 days, nine days ahead of the 60-day statutory target.

Care Homes: Tech Adoption Meets Public Scepticism

For the care sector, the regulatory backdrop matters enormously, but so does something more basic: whether residents, families and staff actually trust the tools being deployed.

New consumer research published this week by Moneypenny, based on a Censuswide survey of 2,000 UK adults conducted between 8 and 10 June 2026, found that 40% of respondents would be happy using AI to obtain general information from a care home such as opening hours. Confidence drops sharply when queries become more personal: just 18% would use AI to ask about a loved one's health or wellbeing, and a third said they would not use AI for any care home communications at all.

That trust deficit sits alongside genuine clinical momentum. Care home providers are deploying AI for fall prevention, overnight resident monitoring and medication management. AI-driven fall prevention is among the most significant advances, with NHS England having announced a nationwide rollout of a predictive tool in early 2025. Firms like Ally Cares, named Most Innovative Care Home Technology Company in the 2026 Global Excellence Awards for the third consecutive year, report that providers using their overnight AI monitoring have seen measurable reductions in falls, ambulance callouts and hospital admissions.

By late 2025, over 80% of adult social care providers had adopted Digital Social Care Records. Skills for Care data puts annual workforce turnover across adult social care at around a quarter of the entire workforce, a crisis that AI-enabled automation is being asked, in part, to address by reducing repetitive documentation and improving governance visibility.

Signals From the Market

Companies House data tells its own story about where investor and founder confidence currently sits. According to AI Business Dispatch analysis of Companies House data, just 17 new SIC 86.10 (hospital activities) companies were incorporated in Q3 2026, a collapse of 94.6% compared with the prior period. The figure is striking. It may reflect founders pivoting away from pure hospital-facing models toward community, primary care and care home services where regulatory complexity is lower and deployment routes clearer. Or it may simply reflect the uncertainty of operating ahead of the MHRA's framework publication.

Either way, the timing is awkward. The regulatory architecture is genuinely improving: faster approvals, fee waivers, live sandboxes, international reliance pathways. But the evidence bar remains high, as the NICE diagnostics advisory committee demonstrated earlier this year when it declined to recommend routine NHS funding for AI-assisted echocardiography tools, despite their reducing cardiac image analysis time from around 550 seconds to 3.2 seconds. Retrospective, non-UK-based validation studies simply were not sufficient.

The sector heads into the second half of 2026 with a clearer regulatory destination than it has had in years, but a route map still being drawn. For care homes, the question is less about what the technology can do and more about whether the people it serves are ready for it.