MHRA - UK Medicines and Healthcare products Regulatory Agency AI Airlock Sandbox
From regulation to collaboration in healthcare innovation
By the Datasphere Initiative | January, 2026
Geographic Scope: National (United Kingdom)
Topic/Sector: Health
Sandbox Type: Regulatory
Year started: 2024
Sandbox Status: In Operation
№ of Applications so far: Pilot phase – 36 applications; phase 2 – 51 applications
№ Accepted: Pilot phase – 4 candidates, phase 2 – 7 candidates
Duration of the Sandbox: 6 months
Who Is Behind This Sandbox: UK Medicines and Healthcare products Regulatory Agency (MHRA)
Overview of UK MHRA's AI Airlock Sandbox
The AI Airlock is a regulatory sandbox launched by the UK Medicines and Healthcare products Regulatory Agency (MHRA) to address the unique regulatory challenges of Artificial Intelligence as a Medical Device (AIaMD). It provides a safe, structured environment, spanning from simulation and virtual testing to real-world deployment, where innovators and regulators can collaboratively explore and mitigate risks while maintaining patient safety.
Why MHRA created the sandbox
The rapid rise of AI-powered medical technologies in the UK, particularly AI as a Medical Device (AIaMD) and other AI-driven digital tools, many of which sit at the borderline of medical device regulation, exposed key limitations in the UK’s 2002 medical device regulations, which were not designed for adaptive learning systems. Traditional approval processes risked slowing innovation and widening health disparities.
To bridge these gaps, MHRA established the AI Airlock Sandbox to:
- Test new regulatory approaches through real-world case studies.
- Identify and generate evidence to inform updated legislation and guidance.
- Support innovation while upholding world-leading standards of safety and efficacy.
These efforts build on the UK’s strong foundation of legislation, guidance, and standards, while aligning the Great Britain regulatory framework with emerging international best practices.
How the sandbox works
The sandbox operates through cohorts of innovators working closely with MHRA case managers to address specific regulatory challenges. Activities span three modes:
- Simulation/roundtables: thought experiments and multi-stakeholder workshops.
- Virtual airlock: testing with real or synthetic data in research environments.
- Real-world airlock: deployment with continuous oversight and safety checks.
Recruitment occurred through an open call, amplified via stakeholder networks, trade associations, and LinkedIn campaigns to attract diverse applicants.
Phase one featured four case studies, combining simulation and real-world testing, while phase two evolved into a challenge-led model structured around three priority regulatory challenges, broadening participation and scalability of learning outcomes.
Challenges addressed
The AI Airlock sandbox tackled several critical challenges in advancing AI regulation for healthcare:
- Defining and maintaining scope, as diverse stakeholder expectations created risks of scope creep in a complex regulatory landscape.
- Timing and funding constraints, which limited opportunities for structured long-term planning and required rapid delivery within tight financial-year cycles.
- Integration of lessons into policy and guidance, with ongoing efforts to embed sandbox insights into “business as usual” regulatory processes.
- Balancing innovation and oversight, ensuring that accelerated testing and collaboration did not compromise patient safety or regulatory rigor.
Results & insights
The AI Airlock sandbox has enabled innovators and regulators to test new ideas in practice, revealing both the technical potential of AI and key regulatory lessons for safer, more effective healthcare innovation.
Philips Medical Systems’ Generative AI in radiology reports
Philips tested a generative AI feature within its imaging system to automate the creation of radiology summaries. Working directly with regulators and radiologists helped refine testing strategies, improve explainability, and strengthen collaboration between R&D and oversight teams.
OncoFlow’s AI for personalised cancer care
OncoFlow developed an AI platform to support cancer treatment decisions by analyzing clinical notes and guidelines. Through the sandbox, the team validated its approach in simulated clinical settings and refined its explainability tools, ensuring transparency and usability for clinicians working with AI-driven systems.
What stakeholders say
Authored by the Datasphere Initiative (2026), this overview captures the core insights of our research. For the full narrative, including comprehensive data and contributor acknowledgments, please download the complete case study report.
Access the Airlock case study
Get practical insights into MHRA’s AI Airlock sandbox, including lessons learned, impact, and what it takes to run a successful AI sandbox.
Enter your email to receive the full case study and related insights.