Zaid Al-Fagih, co-founder and chief executive of AI-powered virtual assistant Rhazes AI, says that if the NHS is serious about AI, it must be placed at the heart of investment planning.
A flagship NHS AI programme was recently paused after 57 million patient records were accessed without consent. The kind of headline that sends ripples through healthcare, government and tech. After all, if the NHS can’t approve its own AI pilots, startups have little hope of scaling their innovations.
But for those of us building AI tools inside the NHS, it wasn’t a surprise; it was inevitable. Complex data rules, slow procurement, and fragmented digital infrastructure leave the UK – and its many startups – stuck in pilot mode. Promising tech clears trials and solves real problems, yet stalls in the NHS maze.
The UK’s healthtech sector is one of the most dynamic in the world. Last year alone, it attracted £27.4 billion in investment. But behind the headlines lies a sector increasingly strangled by red tape, high compliance costs, and a public infrastructure that isn’t ready to deploy what it builds. If nothing changes, we won’t just see startups fail, we’ll see them walk.
Unclear policies
Every AI product I’ve implemented has faced unclear policies, local procurement rules, and missing infrastructure. Startups can spend years navigating procurement at just one NHS trust, while others encounter different frameworks across regions without central guidance on risk or standards. For many, the NHS remains a patchwork of digital maturity, governance, and procurement rules, lacking a single front door for innovation.
The healthcare entrepreneur Ali Parsa has blamed NHS technophobes for stalling innovation. And while he may or may not be right to point the finger at NHS bosses, there’s an even bigger problem at play: NHS infrastructure simply isn’t ready for modern AI, with a leading scientist labelling it “devastatingly user-unfriendly”.
For AI applications to work, it’s essential that health records can be accessed from one hospital to another. But instead, the data is stored at each individual hospital, effectively siloing records within its own IT system. This turns even basic AI integrations into multi-year slogs.
It’s not just infrastructure holding innovators back – regulation and compliance are massive blockers. Data protection and patient safety are absolutely non-negotiable, but the inconsistency, unpredictability and opaqueness of UK rules are pushing founders to the brink.
NHS England decreed that all AI scribes generating clinical notes must be registered as at least Class I medical devices – which typically includes items like bandages and surgical instruments – even if the scribes use well-known large language models (LLMs) like ChatGPT. This effectively forbids doctors from using AI assistants for notetaking unless those tools undergo a full medical device certification.
But frontline clinicians already use general AI tools like ChatGPT informally. One survey found that about one in five UK doctors admit to using unregistered tools. Yet when a company packages the very same technology specifically for clinical use, it’s hit with the highest regulatory barriers.
Doctors and surgeons face mountains of paperwork, and more patients to see in their working day. And even in the short time that I’ve stopped practising, these pressures have grown. So, for me, it seems very contradictory to hit AI scribes with catch-all regulations. It’s discouraging innovation in tools that could save clinicians time and improve patient care.
Deploying abroad
The consequences of all this are now visible. We’re seeing companies pull back from the NHS entirely, looking instead to more supportive ecosystems abroad.
One UK-based medical VR firm, Oxford Medical Simulation, even relocated to the US after years of slow approvals in Britain – a vote of no confidence in our current system. And Hinge Health moved from London to San Francisco, where it grew to a $2.5 billion valuation. Startups are quietly shifting operations abroad when it becomes too hard to deploy healthtech solutions in the UK.
The US is the traditional escape route, but the Gulf states, Qatar, the UAE and Saudi Arabia, are becoming increasingly attractive. I know this because Rhazes AI is one of the companies they’ve welcomed.
Deploying in Doha was quicker, cheaper, and better supported. In the UAE, national AI deployment is similarly advanced. The Ministry of Health and Prevention (MoHAP) has created a dedicated AI Office to govern national rollouts, track outcomes, and integrate predictive models into everyday care.
This kind of national coordination, from sandbox testing to ethics frameworks, is what the UK lacks. If the UK is serious about making the NHS a 21st-century health system, AI must be placed at the heart of investment planning, not tacked on as a soundbite or afterthought.
I’m confident the UK is making inroads. The Medicines and Healthcare products Regulatory Agency (MHRA) recently announced that it will lead a global network on safe AI use in healthcare. And the chancellor has launched a digital transformation fund for public services.
But at the centre of any new initiative, we need clear, streamlined and repeatable pathways for approving AI tools, mandates to make NHS IT systems interoperable, and new, easily accessible sandbox environments where clinicians can experiment with AI safely. Otherwise, we’ll see more talent and investment walk away, not because the UK isn’t brilliant at building, but because it’s failing at deploying.
It’s not just tech talent that’s a flight risk. For healthcare staff drowning in paperwork, patients enduring months-long waits for diagnoses, overstretched hospitals nearing breaking point, and frontline staff battling burnout, AI isn’t a luxury – it’s a lifeline. Time-strained doctors need AI assistants, predictive models, and smart automation to help cut through growing administrative tasks.
Ultimately, AI tools won’t magically work just because talented British teams have built them. They need a system that supports them. And right now, the UK isn’t offering that support. We have a choice to make: either build the 21st-century infrastructure and governance needed to deploy medical AI at scale or keep watching as the future of healthcare innovation moves somewhere else.
