Gemma Warren, Policy Manager, Data and AI
Artificial intelligence is moving quickly in health and care - and for organisations like the HRA, that raises an important question: how do we make sure innovation is happening safely, ethically and with patients at the centre? That’s one of the reasons we’ve been closely engaging with the MHRA’s AI Airlock over the past two years.
The Airlock is a space where regulators and developers can come together to work through some of the trickiest questions around AI in healthcare. It’s deliberately designed as a kind of ‘safe testing ground’ - somewhere you can explore emerging technologies, challenge assumptions and develop regulatory thinking before things are widely deployed.
For us at the HRA, it’s been a valuable way of understanding what’s happening on the ground.
Seeing AI development up close
One of the biggest benefits of engaging with the Airlock has been the opportunity to see how AI tools are being developed in practice - often at a very early stage. That’s important because many of the developers involved aren’t necessarily thinking about their work as ‘research’. They’re building, testing and refining products, sometimes using patient data or aiming for clinical use, but without always being clear on when they might need to engage with processes like ethical review. From our perspective, that creates a bit of a grey area.
A lot of what’s happening in AI development sits right on the boundary of research and that boundary isn’t always obvious. One of the things we’re working on as an organisation is making that clearer, including through our “Is it research?” work, so people know when to come to us and what support they might need. Conversations through the Airlock, and other key initiatives have helped us understand where those dividing lines really are.
It’s not always clear when the scope changes
Something that has come through strongly - particularly in workshops earlier this year - is how difficult it can be to categorise AI systems in healthcare. We’ve looked at examples like clinical note-taking tools or chatbots supporting weight loss and wellbeing. It is not always clear when a product moves from administrative and wellness support to regulated medical use.
The way a system is intended to provide information, what it includes and emphasises and how it frames a response can change what the output is meant to be used for and how it is understood, and can start to shift something from being a general tool into something that looks much more like a medically intended purpose. That matters because once something is performing a clinical function, the expectations around evidence, safety and oversight change significantly.
What’s become clear is that existing definitions don’t always neatly fit AI, especially when tools evolve over time or take on new functions. That’s exactly the kind of challenge the Airlock is designed to explore, and one that the HRA has also been exploring.
Why this matters for the HRA
While much of the Airlock focuses on medical device regulation, there are clear overlaps with our role. Where AI development is conducted to develop, validate, and test clinical hypotheses we’re interested in making sure that robust, ethical research is conducted in a way that people can trust.
But we’re also conscious that AI development can be a long process, developed iteratively as a proof of concept. That’s why engaging early - before something becomes a formal trial - is so valuable for us. It helps us think about when AI development becomes research, and how we can make it easier for developers to engage with the right requirements at the right time.
Alongside this, we’re also seeing AI increasingly used within research itself - whether that’s analysing data or supporting researchers. That brings its own set of questions around transparency and governance, which we’re starting to explore.
Looking ahead
The next phase of the Airlock will launch in Autumn 2026 and aim to continue working with innovators of AI medical devices to further explore the most complex regulatory challenges presented by these tools and systems.
That practical learning is also helping to inform the wider work of the independent National Commission into the Regulation of AI in Healthcare. Established by the MHRA, the Commission is an expert, non-statutory advisory body that brings together global AI leaders, clinicians and regulators to advise the Government on the development of a new regulatory framework for AI in healthcare. In June, the MHRA published the Research and Engagement findings, which are helping to inform the work of the Commission. This brought together perspectives from patients and the public, healthcare professionals, industry and health system partners, alongside insights from the AI Airlock programme. The Commission’s final recommendations are due to be published later this summer.
The MHRA is also aiming to launch a new sandbox initiative focusing on digital mental health, which brings these issues even closer to patients. In that space, AI tools may interact directly with individuals - sometimes in sensitive or high-risk situations. That raises important questions about patient dignity and safety in the conduct of research. It’s exactly the kind of area where we need strong alignment between regulatory approaches and ethical oversight.
Earlier this year, we published our plan to enable safe AI-powered innovation in health and social care research, which sets out how we intend to provide clearer guidance, improve our processes, and support researchers and developers to navigate the system. Initiatives like the Airlock are an important part of that picture. They give us insight into what’s changing and where the challenges are emerging.
AI is evolving quickly, and many of the answers are still being worked out. What’s been really encouraging about the AI Airlock is the openness - the willingness across regulators, developers and others to come together and explore these questions early.
For the HRA, being part of that conversation is essential. It helps ensure that as innovation moves forward, it does so in a way that remains grounded in safety, ethics and public trust.