AI Sweden has signed a research partnership with Genomics England, a company owned by the United Kingdom's Department of Health and Social Care. The agreement provides AI Sweden with unique access to one of the world's leading genomics research environments, meaning the Swedish-developed AI tool, LeakPro, will now be tested to help identify and manage privacy risks in the application of AI within a major genomics library.
The partnership will enable AI Sweden and Genomics England to evaluate how privacy-risk assessments can support the responsible sharing of AI models trained on sensitive genomic data.
![]()
The fact that a world-leading organization like Genomics England is choosing to collaborate with us and test LeakPro in its environment is a fantastic testament to AI Sweden's technical excellence. It also demonstrates how crucial privacy-preserving AI is for the future of healthcare. Together, we can show how AI models trained on large-scale research data can be shared without compromising patient privacy.
![]()
Mats Nordlund (Ph.D.)
Director of Research and Innovation at AI Sweden
Genomics England is a global pioneer in giving approved researchers from approved institutions secure access to deidentified genomic and health data. This is accomplished through a Trusted Research Environment (TRE). There are strict limitations on what can be extracted by researchers, with proposed extractions, for example analyses needed for publication in research articles, needing to pass a review process known as the Airlock. LeakPro will be tested as a tool to support that review.
The project benefits both sides. Genomics England evaluates whether LeakPro can be integrated into its existing governance processes and identifies useful indicators of privacy risk. While AI Sweden gets to test the tool on real models trained on genomic data, which is rarely possible outside a setting like this.
The collaboration is not only an acknowledgment of Swedish research and innovation, but it also opens doors to strengthen Sweden's own infrastructure. The UK has made significant progress in turning technical capability into practical, routine access to data for research.
Fazeleh Hoseini (Ph.D.), a Research Scientist at AI Sweden and lead of the LeakPro project sees the partnership as a chance to learn from an organisation that has invested heavily in secure research infrastructure and is already generating significant value from it at national scale.
![]()
When a model is trained on sensitive data, someone has to decide whether it can be shared. Testing LeakPro in Genomics England's review process shows how measured privacy risk can support that judgement, and gives Swedish regions and data holders a concrete example to build on.
![]()
Fazeleh Hoseini (Ph.D.)
Research Scientist at AI Sweden and lead of the LeakPro project
Francisco Azuaje, Director of Bioinformatics at Genomics England, says:
As AI becomes increasingly important in health research, accelerating discovery and unlocking new insights from data, ensuring it is developed and used responsibly, without compromising participant privacy, is essential.“We are looking forward to working with AI Sweden to help strengthen privacy protections, manage risk, and help researchers using genomic data to drive health breakthroughs. The ultimate goal is to continue building trust while enabling excellent research.
Francisco Azuaje
Director of Bioinformatics at Genomics England
LeakPro is a research project led by AI Sweden and developed in collaboration with organizations in both the private and public sectors. The project is co-funded by Vinnova under the Advanced Digitalization program.
The project develops an open-source framework that quantifies not only the risk of a model leaking sensitive data, but also the impact such a leak could have. In practice,LeakPro works as a stress test: it runs simulated privacy attacks on AI models, including federated learning, and on synthetic data to see whether they reveal information about individuals in the training data.
The tool translates these technical vulnerabilities into concrete risk assessments, helping organizations that handle sensitive data and models meet strict legal requirements such as the GDPR and the EU AI Act.
Related articles