If you want to go fast, you need a good engine. If you want to survive the journey, you also need good brakes.

The AI transformation of healthcare is in progress. We are driven by the hope of faster, highly accurate diagnostics, enhanced access and convenience of care, increased system efficiency and reduced costs. But in healthcare we can’t just ‘go fast and break things’. Mistakes really matter. If AI makes a few wild recommendations on my Spotify playlist, that’s ok. If it generates erroneous diagnoses or dangerous treatment plans, that’s not ok. We don’t need an AI technology to be perfect to be useful, but we do need the bar to be high, just as we would expect from the current standard of care. We need reliable evidence of how good it is, when it might fail, how to detect and mitigate such failures, and how to balance risk and benefit.

This is not unique to AI. With any health technology we want the innovation and adoption pipeline and the regulatory processes within it to be efficient and sufficient. We want to accelerate beneficial innovation to patients and the health system, whilst keeping out products that are unsafe, inequitable, poor value, or simply not ready yet. We want to go fast, but we also need to be able to slow down for the corners.

This happens to be my scientific discipline: the science of innovation and regulation. Within this, my particular expertise is in how it applies to AI in healthcare. The research teams I lead study the innovation and adoption pipeline and the regulatory gates and processes within it. Our goal is that patients and the wider health system get faster access to beneficial AI health technologies (safe, cost-effective, equitable, sustainable), whilst protecting society from those that are harmful. Whilst routine evaluation and regulation is focused on a product, our role is to study the process, to bring the scientific lens to understand, evaluate and optimise it. We create positive impact by ensuring that policy-makers across government, regulators, and the health system have the evidence and insights they need to shape policy and make better-decisions, including in fast-moving, highly complexity areas such as AI.

Normally those of us who work in the science of regulation in healthcare, are working incrementally: understanding, evaluating and fine-tuning one part of that complex system. However sometimes one has the opportunity to re-imagine the whole system in response to a new opportunity or challenge. That has been my privilege for the last year, as I chaired the UK’s National Commission into the Regulation of AI in Healthcare. This initiative was commissioned by the MHRA to advise them and wider Government on what a future regulatory framework for AI in healthcare should look like. It has been characterised by exceptional collaboration across sectors, across healthcare specialities and across diverse scientific disciplines, drawing upon evidence and experts from around the world. 

The Commission heard from more than 12,000 people across surveys, deep public dialogues, expert workshops, and a national call for evidence (to which the Academy formally contributed). The Commission published its Final Report and Recommendations in September 2026. In October 2026 the Government accepted all 44 of the Commission’s recommendations with the Minister for Health Innovation noting “We look forward to taking forward the recommendations of the National Commission and ensuring that the UK remains at the forefront of safe and responsible AI innovation in healthcare.” 

The Commission sought to shape AI policy that is ready for the needs of today and the opportunities of tomorrow. As the Commission concludes, one of my many reflections is on the role of evidence and – with an eye to the scope of the Academy – the importance not just of policy for science, but science for policy.

One area where this feels particularly important is understanding how AI technologies perform once they are deployed in real-world settings. Much of the discussion around AI rightly focuses on innovation, evaluation and regulation before adoption. But approval is not the end of the story. Healthcare systems, patient populations and technologies all evolve over time. Ensuring that AI tools continue to deliver benefit safely, effectively and equitably in practice will require robust evidence, ongoing monitoring and a clear understanding of who is responsible for identifying and responding to emerging issues. These questions will be an increasingly important part of the next phase of AI adoption in healthcare and are an area where organisations like the Academy of Medical Sciences can help convene expertise, evidence and perspectives from across the system.

Returning to our journey: science can help us tune the engine, and sharpen the brakes. It can help us move faster, but it can also help us move forward safely. The scientific community has a key role in shaping this future, not just in the discoveries on which such technologies are built, but in building the system that ensures that these technologies are safe for use, and do what we need them to do. That role does not end when a technology is approved, but continues through generating the evidence needed to understand how it performs in real-world use. Innovation needs regulation, and both of them need science.

Professor Alastair Denniston FMedSci is Chair of Regulatory Science and Innovation at the University of Birmingham where he leads the Bright Group. He also leads the CERSI-AI, the Centre of Excellence for Regulatory Science in AI & Digital Health, and is Hon Consultant Ophthalmologist at University Hospitals Birmingham NHS Foundation Trust. He was the Chair of the UK Government’s National Commission into the Regulation of AI in Healthcare which ran from September 2025 to September 2026.

Related content