| 4 March 2026
There is a very real dilemma facing healthcare organisations across the UK today.
On the one hand, the Government and NHS are increasingly signalling that Generative AI* – having already demonstrated compelling and tangible value – will play an integral role in the future of healthcare delivery. The NHS 10-Year Plan outlines a long-term direction of travel in which AI tools support clinicians in their daily work and help address workforce pressures, enabling more efficient and sustainable care.
On the other hand, regulatory guidance sends a far more cautious message. The CQC advises that AI is “not recommended” where it could influence clinical decision-making, and existing legislation creates significant barriers to compliance for general-purpose consumer GenAI chatbots (such as Claude, ChatGPT and Copilot) in those contexts.
How can a healthcare organisation move forward within this tangled web of guidance?
Leaders across the system are weighing up their options: a 10-year strategy in one hand, regulatory caution in the other, and legislation written long before modern AI tools emerged sitting stubbornly in between.
If GenAI will form part of the long-term future of healthcare, perhaps the prudent approach is to begin adopting it now – building familiarity, testing governance processes, and ensuring staff are not left scrambling when adoption becomes unavoidable.
Yet the laws, guidance and precedent surrounding GenAI adoption remain complex and evolving. It is therefore understandable that some organisations may choose to wait for greater clarity rather than expose themselves to regulatory risk – particularly when the legislation governing many medical applications predates modern AI systems by two decades.
Before considering solutions, we need to understand why this web feels so tangled.
Untangling the web
The UK Government has expressed clear enthusiasm for adopting GenAI across industries, and healthcare is no exception.
Promising findings are also emerging from early studies. A recent literature review found that GenAI chatbots identified certain pathology errors at rates comparable to specialists (89.5% vs 88.5%), classified skin lesions at almost the same levels of accuracy as dermatologists (84.8% vs 84.6%), and staged ovarian cancer at high levels of accuracy in controlled environments (97% vs 88% for radiologists).
It is therefore not surprising that the NHS 10-year plan suggests GenAI could become “every nurse’s and doctor’s trusted assistant […] supporting them in decision making”. Recent commentary indicates that such applications are not merely a hazy goal on a distant horizon, but areas where cautious adoption may begin today. Given the NHS’ influence on sector standards, these signals are likely to shape adoption in the independent sector as well.
So far, the direction of travel appears clear. But strategy documents do not answer the operational question currently troubling many healthcare organisations:
“How do we use GenAI today?”
It is here that the tangle becomes most apparent. While national strategy gestures toward long-term integration, regulatory bodies such as the CQC and various ICBs urge caution where use of GenAI may stray beyond administrative support and explicitly warning that it is “not recommended” for “diagnosis and treatment planning” – the very “decision-making” use cases that the NHS 10-year plan implies will become the norm.
The largest study to date on GenAI use in UK general practice suggests this inconsistency is not only present in national policy, but in local interpretation. Some ICBs have enabled experimentation; others approach with caution. One GP described the experience:
“Our ICB seems so reluctant to accept that AI can be such a useful tool for GPs. I tried to approach them, and was sent an Excel information governance spreadsheet to complete that would need a Master’s in IT to complete – even the AI representative could not fathom it.”
While policy pushes the NHS in one direction, the CQC and local commissioning decisions spiral in another.
This conflict is frustrating for healthcare professionals and organisations, but understandable in context. Regulators and ICBs are operating in uncharted territory asked to provide clarity on rapidly evolving technology with vast and varied use cases depending on the question posed. In such circumstances, it is perhaps unsurprising that the clearest guidance so far – from the CQC – has been to pump the brakes on anything beyond administrative support.
After all, the question “How do we adopt GenAI today?” inevitably leads to further questions:
- Who is responsible if something goes wrong?
- Where is the line between administrative support and clinical decision support?
- What constitutes meaningful human oversight?
- How do we train staff in a way that remains relevant as the technology evolves?
GenAI may simply be too new – and too flexible – for regulators to offer definitive, scenario-by-scenario guidance at this stage.
The red herring of regulatory clarity
In search of certainty, attention often turns to legal reform – particularly to forthcoming Medicines and Healthcare products Regulatory Agency (MHRA) guidance. And there is one paper sitting on healthcare leaders’ desks which has become the centre of attention…
The Medical Devices Regulation (MDR) 2002 requires that any medical device influencing patient treatment must be demonstrably safe and subject to ongoing oversight. However, this legislation was written 20 years before the release of the first widely used GenAI chatbot (ChatGPT 3.5), and its age shows. It does not clearly define the “intended purpose” of a tool that has billions of potential use cases, nor does it easily map onto software that learns and evolves over time.
This issue is being taken seriously by regulators. The MHRA has commissioned industry experts to develop updated frameworks, including secondary legislation which clarifies how MDR applies to GenAI tools. Several “Airlock” reports have explored how regulation might adapt, and updated recommendations are expected later in 2026.
It is understandable, then, that the sentiment among many clinical providers seems to be, “hang tight, MHRA guidance will save us soon”. But while this work is welcome and necessary, it is unlikely to resolve every operational dilemma that healthcare organisations face in relation to AI use.
Future guidance may clarify liability structures and regulatory expectations for certain clinical uses, but it simply can’t answer every governance question, nor remove the need for organisational judgement. Many practical challenges – such as shadow use, staff capability, and proportionate oversight – sit outside the scope of MDR reform yet remain part of the wider web organisations must navigate.
Shades of grey – there’s more than two paths
Presenting this as a binary choice between immediate adoption and complete restraint misses the nuance. Although the tension is real, healthcare organisations are not limited to two opposing options. There are more proportionate paths forward.
Developing an AI policy that establishes core best practice, sets clear boundaries and supports GDPR compliance does not require waiting for regulatory reform – nor does introducing foundational training on appropriate, low-risk uses of GenAI.
Another practical step is acknowledging shadow use. Whether formally endorsed or not, many clinicians and non-clinical staff are already experimenting with consumer GenAI tools to reduce workload pressure – with many finding them incredibly helpful. Providing guidance on safe drafting, documentation support and workflow enhancement – while clearly highlighting areas that cross into regulated clinical territory – is a measured response to stepping out of the grey and into the clear.
These are not radical transformations. They are structured first steps which organisations can begin taking immediately.
Back to the desk: what should we actually do?
For the healthcare leaders frowning at their desks, two things have become clear.
First, the anticipated 2026 regulatory updates will not resolve every uncertainty organisations face today. Second, the dilemma is not as black and white as it may initially seem.
The more useful question is not: “Should we do nothing until clarity arrives, or begin a full-scale GenAI rollout today?” But rather: “What proportionate steps towards GenAI use make sense for our organisation right now?”
Either way, GenAI will form part of UK healthcare in the long term. Choosing to delay entirely is a decision in itself – one that may allow overworked staff to continue experimenting without oversight and compress adoption into a future moment of urgency.
Attempting to address every aspect of GenAI adoption at once would be unrealistic. But taking modest, structured steps today is both possible and prudent.
The sensible path sits between hesitation and haste, avoiding the two extremes of waiting in fear or moving too quickly. Structured guidance and practical training can help organisations take those first steps safely, proportionately and with confidence.
*Generative AI (GenAI) is a branch of artificial intelligence which produces new content and original data in response to prompts.
Next steps
Are you exploring the safe use of Generative AI in healthcare?
Explore our eLearning course GenAI in Healthcare: Starting Safely.
Find out more