PHASE IV AI Final Conference Explored the Future of Privacy‑Compliant Health Data and AI in Europe

The PHASE IV AI Final Conference was held in Turku on 4 June 2026 as a hybrid event. Antti Airola (University of Turku) opened the event by introducing the consortium and outlining the project’s aims to advance the use of privacy-compliant health data for artificial intelligence (AI) in Europe.

From project results to policy impact: Advancing health data and AI in Europe

In the opening remarks, Nihal Yildirim (HaDEA) described PHASE IV AI as an ambitious Horizon Europe project that has delivered strong scientific and technological outcomes. The project has developed a federated learning infrastructure, a synthetic data framework, and validated solutions in three clinical use cases: lung cancer, prostate cancer, and ischemic stroke. These results demonstrate that advanced AI can be developed and deployed while maintaining strict data protection requirements. The project is part of the HEALTHDATA4EU cluster of six initiatives.

In her keynote, Christina Kyriakopoulou (European Commission, DG RTD) outlined how the Data Act (2023), AI Act (2024), and the European Health Data Space (EHDS, 2025) are shaping the future of health data use in Europe. These initiatives aim to enable secure, interoperable, and scalable use of health data across Europe. At the same time, they introduce new requirements for data governance, interoperability, and compliance, which projects like PHASE IV AI must address.

PHASE IV AI Health Data Hub

Geert Machtelinckx (Fujitsu) introduced the PHASE IV AI Health Data Hub, a governance framework aligned with EHDS. The Hub enables secure data exchange without centralising sensitive data through “data visiting” approaches, where computation is brought to the data rather than moving data across systems. The Health Data Hub combines federated learning, synthetic data, and secure processing environments with identity management and data catalogues, helping address the challenge of fragmented health data across institutions.

Clinical impact through real-world piloting

The project’s clinical relevance was highlighted in the presentation by David Brown (Nottingham Trent University), focusing on the piloting activities. In lung cancer, AI supports earlier detection through EHR-based risk prediction and advanced CT imaging, including synthetic image generation. In prostate cancer, federated models improve decision-making without the need for invasive procedures. In stroke care, AI supports faster and more effective treatment decisions. These use cases demonstrate how federated learning and synthetic data enable collaboration across institutions without sharing sensitive patient data. Building clinical trust through validated and explainable AI remains essential for adoption.

Synthetic data as a key enabler

Rafael Redondo (Eurecat) and Ileana Montoya Perez (University of Turku) presented advances in synthetic data. Redondo highlighted LAND (Lung and Nodule Diffusion) as an example of diffusion-based image generation that improves realism in medical imaging. Montoya Perez emphasised that synthetic data must be evaluated based on quality, privacy, and utility, ensuring it is realistic, safe, and useful for tasks such as training AI models. The project has also developed tools such as SynGPipe to support this work.

Privacy-enhancing technologies (PETs)

Mariya Georgieva (Tune Insight SA) introduced the privacy-enhancing technologies underpinning the project. These include federated learning, homomorphic encryption, secure multi-party computation, and differential privacy. Together, these methods enable analysis of sensitive data without exposing it, ensuring protection during both computation and result generation. Combining multiple PETs is necessary to balance privacy, performance, and scalability in real-world healthcare environments.

Regulatory, ethical, and practical challenges

The afternoon session focused on regulatory, ethical, and practical challenges related to the implementation of EHDS which were further explored in the panel discussion.

Building trust: safe and secure data use

Peija Haaramo (Findata) described how the EHDS will expand Finland’s centralised data access model into a European cross-border system. She highlighted the need for coordinated infrastructure, stronger requirements for data quality and transparency, and expanded roles for data access bodies. Safe data use relies on legal, technical, and social trust, supported by clear governance, secure systems, and public acceptance. Haaramo also identified key success factors, including sufficient resources, clear EU-level guidance, and strong collaboration between stakeholders across countries.

Ethical bottlenecks in cross-border data access

Jenina Soimala (THL) emphasised that ethical governance remains largely national, creating fragmentation and delays in cross-border research. Differences in ethical review requirements, administrative complexity, and inconsistent interpretations—such as the distinction between research and quality assurance—can significantly slow down projects. She highlighted the need for greater harmonisation, clearer guidance, and stronger collaboration between national ethics bodies.

Industry perspective: scaling data use

Tuomo Pentikäinen (VEIL.AI) provided an industry perspective on the growing demand for privacy-preserving solutions in health data use. He noted that organisations are increasingly using data for AI-based diagnostics, real-world evidence, and model development, often in multi-country settings. At the same time, regulatory complexity, strict consent requirements, and privacy risks create significant challenges. Pentikäinen presented anonymisation and synthetic data as practical solutions that enable data use while protecting privacy. He stressed the importance of balancing privacy with usability and warned that fragmented national implementations of EHDS could limit Europe’s ability to scale innovation.

Panel discussion: The European Health Data Space (EHDS), technical and legal perspectives

The panel explored the EHDS from technical, regulatory, research, healthcare, and industry perspectives, focusing on its potential to support data driven health technology development while addressing key implementation challenges and possible solutions.

A common message was that EHDS has significant potential to accelerate health research, AI development, innovation, and commercialization across Europe. Shared European practices and harmonized rules were seen as essential for enabling cross-border access to health data and supporting both academia and industry. The main challenges are not technological, but organizational, political, and collaborative.  As Miikka Kallberg (CSC) noted, the real problem is not technical, but the willingness to truly start collaborating.

Panellists agreed that most required technologies already exist, but common standards, harmonized processes, and implementation models are still lacking. The focus should therefore be on defining shared practices rather than developing new technologies. Privacy‑enhancing technologies (PETs), such as federated learning, differential privacy, and homomorphic encryption, were highlighted as increasingly important. The panel noted that EHDS currently places limited emphasis on federated approaches, even though these may be critical for large‑scale cross‑border analytics without moving sensitive data.

The need for trusted and certified technologies was also discussed. As Tuomo Pentikäinen (VEIL.AI) noted: in the future, we may see certification schemes for individuals and organisations that can provide assurance on complex issues such as whether data has been properly anonymized.

Synthetic data emerged  as a key topic. While seen as highly promising for software testing, innovation, AI development, and preliminary research use cases, participants acknowledged its limitations and current legal uncertainty. As Ibrahim Sabra (University of Vienna) stated: at present, the EHDS lacks a clear definition and standards for synthetic data—an important gap that should be addressed in the implementing act.

The panel also highlighted challenges in harmonising interpretations across EU member states. Jenina Soimala (THL) pointed to the Nordic collaboration models as positive examples of how countries can work together through legal, ethical, metadata, and technical working groups to align their approaches.

Another recurring theme was the balance between data protection and innovation. Several speakers underlined that health data should not only be viewed as a regulatory liability, but also as a strategic asset that can support better healthcare, AI, and medical innovation when governed responsibly. As Arho Virkki (VARHA) noted: Public data holders, such as hospitals, are obliged to make data available, but the key question is how to incentivise them to produce high‑quality anonymised data products.

Panel chair, Juha Pajula (VTT), concluded the panel by highlighting four key priorities:

• Europe urgently needs common practices and standardised approaches for EHDS implementation.
• Continuous exchange of best practices between member states must be strengthened.
• Trusted and secure technology frameworks or certification mechanisms should be developed.
• The role of synthetic data within EHDS needs to be clearly defined.

 Closing remarks

Antti Airola concluded that PHASE IV AI has demonstrated that privacy-preserving AI in healthcare is both feasible and impactful. He noted that the project will continue with the final piloting phase running until early October and invited stakeholders to participate in upcoming workshops and collaboration activities. He emphasised that building trust and aligning innovation with regulation will be essential for achieving large-scale impact in Europe.

Further information

Further information is available on the conference website, including:

• Presentations and speaker details
• Poster and demonstration sessions (available as PDFs), showcasing additional project results, including tools supporting EHDS readiness, methods for improving data quality and interoperability, and solutions based on federated analytics and synthetic data

For further information follow the project on LinkedIn.