SDEs play a central role in UK health data policy by providing controlled access to de-identified NHS datasets. However, current SDEs are often not optimised for commercial AI innovation, they can be slow, costly, and restrictive. This project explores how SDEs can be enhanced to support the full AI product lifecycle, from early-stage experimentation using synthetic data to pre-market development and post-market surveillance. It addresses both technical requirements (e.g., scalable compute, smooth ingress and egress of data and code) and governance challenges (e.g., mitigating disclosure risk from model outputs), aiming to create environments that are secure, functional, and trusted by users and the public alike.

• Expert workshops: Conducting national workshops to capture industry needs and public/governance perspectives.
• Technical proposals: Developing design objectives for an AI innovation platform within SDEs.
• Risk framework: Co-developing a pan-SDE model risk assessment framework to manage disclosure risk.
• Policy alignment: Collaborating with DARE UK and regulators to standardise guidance, best practices, and governance for AI healthtech innovation.

This project explores how LLMs should be assessed for qualification as a medical device and risk classified when used for medical purposes. It identifies common healthcare use cases for LLMs, develops evidence-based archetypal intended-use statements, and applies established regulatory principles to determine whether those uses fall within the definition of a medical device. The work will identify the characteristics that influence risk classification and produce policy recommendations for UK regulators (MHRA, NICE, CQC) and international bodies. It also aims to guide the future regulatory landscape by clarifying how LLM-based medical use cases should be assessed and governed, outlining what an optimal future approach to regulating these technologies should entail, and ultimately creating clarity on how LLMs fit within medical device regulation what that means for their safe and effective use.

• Evidence synthesis: Conducting a rapid scoping review of biomedical literature to identify healthcare applications of LLMs.
• Stakeholder engagement: Hosting multi-stakeholder workshops with regulators, clinicians, legal experts and industry to refine intended-use statements and explore risk classification; identify gaps and inconsistencies in how current regulations are applied to LLMs in healthcare; clarify where uncertainty persists; and map how best to shape the regulatory landscape going forward.
• Policy development: Drafting clear, actionable recommendations for MHRA and international regulators.
• Public involvement: Running a PPIE workshop to gather patient and public perspectives on safety, fairness and accountability in LLM deployment.
  Dissemination: Publishing the scoping review, policy report and workshop outputs, with UK-wide and international dissemination via CERSI-AI and the NIHR Incubator for AI and Digital Health.

The AI Readiness Checklist is a practical self-assessment and diagnostic tool designed for healthcare organisations preparing to implement AI technologies. It provides a structured way to evaluate readiness across key domains such as data and digital infrastructure, governance and assurance, workforce capability, ethics and equity considerations, and organisational strategy. By completing the checklist, organisations can identify strengths, highlight gaps, and prioritise where to focus their improvement efforts, supporting more successful, safe and sustainable AI adoption.

We are making this project a reality through extensive research, engagement and real-world collaboration. The checklist has been developed through desk research, expert interviews and stakeholder consultation to ensure it reflects the practical needs of NHS organisations. We are piloting the tool with NHS trusts to gather feedback and refine its usefulness in live settings. In addition, we are working closely with devolved nations, including running a joint workshop with the Scottish Government to explore how the checklist aligns with NHS Scotland structures and priorities. We are also beginning to test how the tool could support primary care and social care settings, recognising that AI adoption is expanding beyond acute providers. By openly publishing the tool on aireadiness.uk and aligning it with national guidance and regulatory expectations, we aim to create a consistent, practical resource that supports safe, effective and scalable AI adoption across the whole health and care system.

HaRi aggregates device registrations, safety data, NHS procurement information, and peer-reviewed evidence from sources like Public Access Registration Data Base (PARD), UK, Food and Drug Administration (FDA), USA, The European Database on Medical Devices (EUDAMED), and PubMed, enabling searches across the UK, Europe, USA, Canada, and Australia.

By unifying and contextualising regulatory intelligence, HaRi empowers safer, faster, and more transparent decision-making across the medical device ecosystem.

We are developing the UK’s first borderline manual specifically for AI as a medical device (AIaMD) and Software as a Medical Device (SAMD). This guidance document addresses the critical question facing AI health innovators: “Is my product a medical device, and if so, what risk class does it fall under?”

The manual aims to provide detailed case studies across multiple clinical domains and each case study demonstrates the decision-making process for both qualification and classification.
Outputs will include consensus recommendations, illustrative examples, and guidance for innovators and regulators.

The manual uses a systematic approach making it easy for manufacturers to find relevant guidance for their specific product type. We have developed archetypal use cases that demonstrate classification progressions, showing how subtle changes in intended use or functionality can shift regulatory status. Multi-stakeholder workshops will be conducted with innovators, clinicians, regulators, and legal experts to build consensus on classification and regulatory approaches. Outputs will be harmonised with the existing EU borderline manual to avoid conflict and confusion, and will be formally submitted to the MHRA as recommendations for official guidance. The manual aims to provide clear, practical advice to innovators on borderline cases, helping them plan development pathways, meet compliance requirements, and accelerate safe adoption of AI health technologies.