1. Background

As the UK’s national institute for health data science, Health Data Research UK (HDR UK) has been central to advancing the trustworthy use of data across research and health systems. Advanced analytics, including Artificial Intelligence (AI) has played a central role in the Institute’s activities from the outset. In the same period, the pace of AI innovation has accelerated and emerged as a transformative force in all sectors including health and biomedical discovery.  

Recent national policy ambitions underscore the growing importance of AI across science and health. The Department for Science, Innovation and Technology (DSIT) AI for Science Strategy highlights the central role of AI in drug development, including target discovery, modelling and early-stage validation, and stresses the need for secure access to high impact datasets to support this work. The DSIT strategy also signals a clear expectation that UK infrastructure should enable responsible use of sensitive data for AI driven discovery research. This strengthens the case for HDR UK to support the safe use of AI in research settings and to create the conditions for trusted data access at scale.

The examples highlighted in the strategy, particularly in AI enabled drug development, reinforce that our work must bridge scientific ambition with robust information governance across trusted research environments. 

How is AI defined in this strategy

Artificial intelligence is the use of computer systems to perform complex tasks usually associated with human intelligence. In the context of HDR UK’s activities, it is helpful to provide a more technical definition that is explicit about what is and is not in scope. In this strategy, AI refers to emerging methods of scientific research that rely on data intensive machine learning systems.

 

This includes approaches that are fundamentally enabled by large scale datasets and modern machine learning techniques, particularly deep neural networks. These methods may also be combined with probabilistic or rule-based components where relevant. Traditional statistical modelling and predictive analytics are not included unless they are based on data intensive machine learning.

AI is reshaping how research teams access, analyse, and apply health data insights—from molecular discovery to provision of real-world evidence for policy change. Yet its safe and effective use depends on high-quality data, ongoing transparency, and strong information governance. HDR UK is uniquely placed to enable these foundations, linking data assets, Trusted Research Environments (TREs), researchers and innovators across universities and industry, and the NHS.

Building on the Institute’s internal AI Landscape assessment (2025) and engagement with experts and stakeholders, this strategy sets out HDR UK’s role as the UK’s trusted enabler of responsible AI in health data research. Stakeholders consulted include the HDR UK Strategy and Integration Group (SInG), HDR UK Senior Leadership Team, HDR UK Public Advisory Board (PAB) and International Advisory Board (IAB). This strategy describes the vision, strategic drivers, objectives, and delivery approach for 2026–27.

HDR UK’s official AI Strategy was approved by our Board of Trustees in December 2025.

2. Context and strategic drivers

The development of the AI Strategy draws on internal review and broad consultation across HDR UK’s community, partners, and funders. It also considers rapid advances in AI and machine learning, growing policy attention to information governance and safety, and the need for coherence across UK research and health data infrastructure.

Key drivers include:

  • National leadership: ensuring HDR UK aligns with NHS/ Government priorities
  • Trust and transparency: strengthening public confidence through established PPIE structures and demonstrating best practice.
  • Scientific opportunity: connecting multimodal datasets for AI-enabled discovery.
  • Infrastructure readiness: sharing frameworks to make TREs AI-ready and reduce duplication.
  • Skills and community: convening academia, NHS, and industry to maximize expertise and share learning.
  • Impact: demonstrating how AI advances science, improves outcomes, and drives efficiency, ultimately accelerating the pathway from discovery to clinical impact.

Through its UK-wide networks, HDR UK brings together the groups that must work in concert for AI to deliver national value. Health systems and data controllers run secure environments and hold operational expertise. Academic and clinical scientists provide methodological depth. Industry partners contribute technical capability, compute, and investment.

HDR UK sits at the centre, helping drive alignment, ensuring that the foundations for safe and effective AI are in place, and supporting teams to move from isolated pilots to solutions that can work across national infrastructure.

 2.1 Existing foundations

The strategy will build on the existing and ongoing activities carried out throughout HDR UK, including the DARE UK programme, which provides the emerging technical foundations needed to support safe and scalable AI in health data research. Relevant existing DARE UK projects include; Semi Automated Checking of Research Outputs and Support for AI (SACRO ML), Federated Research Infrastructure by Data Governance Extension (FRIDGE), and Guidelines and Resources for Artificial Intelligence Model Access from Trusted Research Environments (GRAIMatter).  

The BHF Data Science Centre has also initiated a range of activities to support AI innovators, including working with the NHS England National Secure Data Environment to apply the Foresight model to 57M de-identified patient records.  

Together these projects show what is already working, supply practical reference points for secure AI development, and demonstrate where further alignment and specific attention is required across HDR UK to add value. 

2.2 Cross cutting strategic needs

As AI methods become more embedded in research and service improvement, several system wide needs have become clear. These needs require collaboration across health data owners, trusted research environments, academic groups, and industry partners, and they shape the foundations required for safe and scalable AI. 

First, AI for data curation is increasingly important. Many organisations face growing volumes of complex health data that need to be curated and cleaned before advanced analysis can take place. AI enabled curation can help automate these tasks, support consistency across datasets, and reduce the time it takes to prepare data for secure use. This work requires closer coordination with data controllers and Trusted Research Environment (TRE) teams so that curation methods align with Information Governance (IG) requirements and can operate securely. 

PAIR Project [in progress]

PAIR (Building a Pipeline for utilising foundation AI on EHRs) tackles the challenge of supporting hypertension clinics where clinicians have limited time to review complex records. The project builds a reusable pipeline for applying large language models to routinely collected clinical data within a safe setting by developing a cloneable pipeline that can run in a secure environment. PAIR offers a concrete example of how real-world data, trusted environments, and advanced AI capabilities can be combined to support better research and clinical care.

AI for early drug discovery [to be further developed]

Significant progress has already been achieved applying AI to drug discovery using pre-clinical/non-human data including the AlphaFold protein structure prediction breakthrough, led by Google Deepmind and underpinned by the European Bioinformatics Institute Protein Data Bank. However, the pipeline for drug discovery also requires multi-modal human data for both early phase discovery e.g. multi-omics analysis for target identification and to bridge the gap between preclinical (animal) studies and early phase clinical trials.

 

An AI for drug discovery signature project will leverage HDR UK’s existing expertise in use of deeply characterized consented data to support early stages of drug discovery by working with large molecular datasets and human data. It will require collaboration with consented cohort partners and close industry engagement to address the challenges of deploying a range of AI methodologies within trusted research settings.

Second, training and capacity building remain essential. Many teams are navigating new IG requirements, new methods, and new routes to impact. Adopting a team-science approach—aligned with HDR UK’s commitment to a collaborative, inclusive research culture—means building shared capability across TRE teams, analysts, clinicians, technical specialists, and researchers. This supports consistent practice, improves safety, and reduces reliance on isolated experts by recognising the diverse skills that drive AI innovation and high-quality health data research. Ongoing work is required to provide structured programmes as well as practical materials that help teams collectively understand how to use AI methods responsibly in trusted research environments.

Third, industry engagement requires defined pathways. Technology “hyperscalers” and life sciences companies bring capability, compute and tools that can accelerate development, but they also face constraints when working within UK data and information governance frameworks. Creating a structured route for engagement, aligned to HDR UK’s industry strategy, will help align expectations, reduce duplication, and enable shared work on areas such as secure compute, pipelines and model validation. 

Together, these needs point to the importance of clarity and alignment across the system. They highlight that progress will rely on technical capability, shared practices, clear IG and embedded public involvement, through coordinated engagement across diverse organisations that contribute to the UK’s data landscape. 

HDR UK can enable progress in several ways; by helping data partners prepare for AI, including clarity on good practice for TREs and support for AI for data curation so that large and complex datasets can be organised and used safely. Secondly, by supporting research teams to test methods in structured settings that make clear how AI can move from early exploration to scalable use, backed by stronger training and capacity building. Third, by maintaining open engagement with the public and policymakers so that confidence in the use of AI grows alongside technical capability. Finally, by creating clear pathways for industry engagement and by building connections across active groups so learning is shared and duplication is reduced. 

3. Strategic objectives, delivery and monitoring progress

Overarching objective:

Embed safe, equitable, and trustworthy AI as a core enabler of HDR UK’s mission, ensuring AI readiness across UK data infrastructure, strengthening public trust, and accelerating AI enabled translation from data to improved health outcomes.

It is proposed that the strategic objectives for HDR UK’s AI engagement over 2026–27 are structured to align with the Institute’s overall mission, integrating AI across programmes, and delivering measurable outcomes through four operational pillars. 

1. Scale AI learnings across data infrastructures - ensuring national TREs and data assets are well curated, unbiased and robustly governed to be ready and safe for AI use

2. Build holistic AI research signature projects - demonstrating AI’s value through scalable proof-of-concept projects in real-world health and research settings.

3. Be transparent and demonstrate trustworthy AI - embedding communication, ethics, and public involvement to ensure responsible AI deployment.

4. Align AI communities of practice - reducing duplication by convening focused groups to solve specific problems, share practical methods and produce actionable outputs across NHS, academia and industry.

  • Accelerating AI adoption across TREs by co-creating common guidance and principles shaped by users and aligned with established materials e.g 2021 Health Data Research Alliance paper 2 on TRE good practice, which has been downloaded more 12,000 time and the more recent GRAIMatter, DARE UK paper on recommendations for disclosure control of trained machine learning models from TREs3.  

    • A coordination team, working with the Alliance, will convene TRE operators and data governance leads to co-develop a set of AI readiness guidelines on best-practices when working with AI within TREs, reflecting practical needs across a range of data modalities and research uses.  
    • A draft set of principles will be piloted within selected TREs and refined through user testing and feedback 
    • Learning will be packaged as practical implementation guidance and templates  
    • A shared workspace will be created so that TREs can contribute examples, raise issues, and track adoption 
    • Ongoing reporting will monitor the percentage of TREs using the principles and highlight where further support is needed. 

  • Creating an operational pathway from concept to adoption that shows how AI is used in practice to deliver discovery research and health impact, within secure data infrastructures. 

    • Two to three signature projects will be selected using clear criteria, such as data readiness, public benefit, and potential for scalability and reuse 
    • Each signature project will operate as a structured project with defined milestones, decision gateways, documentation expectations, and evaluation plans 
    • All workflows, methods, technical tools, and governance steps will be fully documented to generate reusable process papers 
    • A pipeline or toolkit from each exemplar will be designed for reuse across other research teams and aligned with Pillar 1 outputs. 
    • Findings and infrastructure requirements will feed into the TRE working group to enable scale across the network 
    • Signature projects will be targeted to exemplify a range of research domain impacts, from drug discovery to increasing clinical pathway efficiency. 

  • Ensuring trust is active, iterative, and visible, rather than an afterthought once technical work is complete. Delivering trust will require ongoing engagement, clear explanations and close attention to fairness and avoiding inequity.  

    • Public and researcher engagement will be embedded throughout activity (e.g., within each signature project) rather than treated as a final step 
    • A rolling calendar of open engagement sessions will be scheduled and integrated with the HDR UK Public Advisory Board 
    • All outputs will include plain language explanations and graphical materials designed with input from PPIE colleagues 
    • Working with the communications team to support creation of web content, social posts, explainers, and short form material for broad audiences 
    • AI safety and transparency updates will be coordinated so that evidence and guidance is always accessible 
    • A strand of work will focus on explaining fairness considerations and how risk of bias is being addressed.  

  • Promoting collaboration and shared learnings to avoid duplication and bringing diverse expertise across AI and health within academic, NHS and commercial shared forums   

    • An external AI landscape map will be developed with capability information, tools in use, pipelines, IG models and ongoing projects to allow others to see what can be done and where.  
    • Using the AI landscape map to drive activities, targeted groups will be identified based on specific outputs and invited into a working community where learning, opportunities, and risks can be shared 
    • Associated “task and finish” groups will be convened to support specific deliverables across all 3 strategic pillars 
    • Building on existing communities of practice to provide an ongoing feedback route for the AI strategy and support alignment with national and international initiatives 

4. Delivery leadership

Ongoing oversight by the Senior Leadership Team and the Board will include quarterly OKR updates.  Implementation and delivery will be coordinated by the HDR UK strategy team under the Director of Strategy, with activities embedded across Institute programmes.  

An AI Advisory team (Dave Robertson and Alastair Denniston) will provide expert leadership and ensure alignment across HDR UK’s and the wider AI ecosystem. The Chief Data Officer will provide additional important technical leadership, working with colleagues across the Institute.