Three takeaways from our 2026 AI in the NHS Summit

Over 2026, hopes that AI can help tackle some of the thorniest problems facing the NHS have continued to grow. The dominant policy questions no longer focus so much on whether AI can or should be used to improve health services, but on how best to deploy AI and assess performance.

In October the Health Foundation hosted its third AI in the NHS Summit. This year’s summit brought together speakers from the UK, US, Singapore and South Korea to look at what we can learn about AI implementation from leading systems, and how the UK can conduct faster, rigorous evaluation. 

Key takeaways included the uneven distribution of the benefits of AI, the difficulty of generating sufficient evidence on AI tools and how the UK can make the most of its assets. 

Pockets of progress

Speakers at the summit discussed several areas where significant progress is being made, both in the adoption of AI – including the increasingly widespread use of ambient voice technology and the growing ease with which clinicians are able to build and deploy their own AI tools to solve specific problems – and on broader policy architecture. For example, in the UK the government recently accepted all 44 recommendations made by the MHRA’s National Commission into the Regulation of AI in Healthcare. 

At the moment, progress on adoption is largely outpacing developments in policy and regulation. In addition, progress is happening in isolated pockets, both geographic – with advancements taking place in the best resourced and most digitally mature areas – and thematic – with AI tools being developed and deployed where data are readily available and adoption is easiest. 

During the summit, Ricardo Baptista-Leite, CEO of HealthAI, described a ‘graveyard of pilots’ not developed with equity or scalability by design, warning that AI readiness funding from governments tends to think ‘let’s make our best hospitals or the ones that are best located in the cities the first ones to be AI ready… so you’re actually widening that gap’. 

The question of how we roll out AI fairly and coherently across a whole system, and what national policymakers need to put in place to make this happen, remains pressing. With the capability and availability of AI tools moving so rapidly, there is a real risk that they outpace our ability to govern and steer them. Policy will need to find a way to catch up, and quickly.

The evidence bottleneck 

Of course, generating the evidence to inform policy decisions is critical. Justifying the cost (both in terms of resources and opportunity) of whole-system deployment requires a high-level of evidence on safety, efficacy and the ability of a tool to integrate and perform well within real-world workflows.

Most health systems have either just begun to think about how they can rigorously evaluate AI at pace or are struggling with how to make it happen in practice. Barriers such as inadequate infrastructure and funding are prevalent. 

Mary Dixon-Woods, Director of THIS Institute, described the AI evaluation landscape in the NHS as ‘the wild west’, where at the moment, ‘It is possible to go straight to large-scale deployment without that testing in real-world contexts being done.’ This was supported by Jennifer Dixon, Chief Executive of the Health Foundation, who highlighted that most NHS trusts lack the time, expertise and funding to conduct appropriate evaluations. The current infrastructure for assessing AI is largely set up for research and assurance, with far less focus on promoting the adoption and spread of beneficial technologies. It is the latter we now need to develop in order to assess the value of AI technologies to the health system, not just the effectiveness of individual innovations.

Even world-leading health care institutions are finding evaluation of AI challenging. Micky Tripathi, Chief AI Implementation Officer at the Mayo Clinic, said that although Mayo is excelling at developing AI, ‘...where we’re faltering is the ability to appropriately vet those solutions as they’re coming through.’ The Mayo Clinic is now beginning to establish its process and governance for rapid cycle review. 

A tension was apparent in these discussions. While speakers agreed that AI models need to be tested locally, it was also clear that for most health care systems, the cost of conducting multiple single-site evaluations is not sustainable. 

Making the most of what we’ve got 

These questions – of equity, resource-distribution and evidence-generation – are daunting, but speakers from outside the UK gave a useful reminder that, in many ways, we are in an enviable position. As a single-payer system with the ability to take a whole-life view of a patient, high-quality health data and a strong research ecosystem, there is more we could do to capitalise on these assets. 

Our work as the research partner to the MHRA's independent National Commission into the Regulation of AI in Healthcare found that the public is largely supportive of the use of AI in health care. That support is conditional on strong safeguards where accuracy and human oversight are essential, so making fast progress on the issues highlighted at the summit will be essential to maintain public support. 

We are still missing a clear national policy architecture to ensure decisions about the use of AI in health care are evidence-based and equitable. The UK is starting to develop some pieces of the puzzle, as seen with the MHRA’s recommendations and various funding pots and programmes, including a new AI Challenge Prize announced in last week’s budget.

The task now is to complete the picture, linking work together across government to ensure policymakers are actively steering the development of AI towards better health and care. One of the priorities must be to test and evaluate new innovations in real-world NHS settings. The Health Foundation’s upcoming work on rapid evaluation, international learnings, payment systems and more will aim to inform this approach.

This content originally featured in our deep dive newsletter.