🔍💡 AI is transforming European health systems: 74% of member states are already using AI for assisted diagnosis, but only 41% have defined legal liability rules in case of error. An essential WHO overview to navigate between innovation and patient protection.
📌 This report is essential for public health officials, institutional decision-makers, and professionals engaged in the digital transformation of health systems. It provides a factual and comparative overview of the integration of AI in 27 EU member states, allowing for the identification of regulatory, ethical, and operational gaps that hinder or facilitate the responsible deployment of these technologies. Concrete data on national strategies, legal frameworks, and applications already deployed offer stakeholders benchmarks to position their initiatives and anticipate regulatory developments, particularly the implementation of the European AI Act and the EHDS by 2029.
This WHO report is essential for prevention stakeholders as it allows them to anticipate the transformation of their practices (74% of countries are already using AI for assisted diagnosis), navigate the complex regulatory framework of the European AI Act (applicable from August 2026), and identify concrete opportunities (chatbots, predictive screening, risk assessment tools) while developing a critical perspective on algorithmic biases and potential inequalities. It reveals major gaps (only 7% of countries have specific ethical guidelines, 18% consult the public, 26% train their professionals) that justify advocacy for participatory and equitable governance. Finally, it prepares stakeholders for the EHDS 2029, identifies available resources (European funding, territorial pooling) and provides a European benchmark to identify transferable best practices to ensure that AI serves health equity rather than reinforcing discrimination.
Source: 📒 Artificial intelligence is reshaping health systems: state of readiness in the European Union
✍️ Artificial intelligence is reshaping health systems: state of readiness across the European Union - WHO Regional Office for Europe — April 2026
📜🔗LINK to the source
Number of pages: 64
1. ANALYTICAL SUMMARY
Context and issues: A regulatory and technological turning point for Europe
The year 2025 marks a decisive moment for the integration of artificial intelligence (AI) into European health systems, with the adoption of the first comprehensive global legislation on AI: the EU AI Act. This report draws on the 2024-2025 WHO survey of 50 member states in the European Region (response rate of 94%), including the 27 EU member states. It examines where countries currently stand in terms of policy frameworks, regulatory approaches, data availability, trends in AI adoption, and encountered obstacles. Health systems are facing major challenges: reducing pressure on professionals (96% of states consider it a major or moderate driver), improving patient care (100% of states), and increasing efficiency (89% of states). AI is seen as a potential response, but its deployment raises crucial questions of ethics, governance, accountability, and patient protection.
Operational contributions: Comparative data to guide public action
The report provides a detailed mapping of national strategies (85% of EU states have an intersectoral strategy on AI, 11% have a health-specific strategy), legal and ethical frameworks (only 7% have issued specific ethical guidelines for health), health data governance (67% have a national strategy on health data, 63% have a national data hub) and applications already in use. It identifies major barriers to adoption: financial accessibility (41% consider it a major barrier), legal uncertainty, and data quality. The document proposes concrete action points organised into eight themes: national strategies, stakeholder engagement, legal and regulatory landscape, data governance, opportunities and applications, barriers and political facilitators. These recommendations enable stakeholders to anticipate future regulatory requirements, particularly alignment with the AI Act and the implementation of the European Health Data Space (EHDS) scheduled for March 2029.
2. KEY POINTS OF THE DOCUMENT
1. Fragmentation of national strategies despite a common dynamic
85% of EU member states (23 out of 27) have implemented an intersectoral national strategy on AI, while 11% (3 out of 27) have a health-specific strategy. Although most countries have adopted a strategy, many are still in the early stages of review or lack a clear definition of AI. The responsibility for oversight and implementation generally rests with existing government agencies, with fewer countries establishing new independent bodies fully dedicated to this. This fragmentation creates challenges for harmonisation and cross-border coordination (pp. 6-9).
2. Limited engagement of patients and the general public in AI governance
81% of EU member states (22 out of 27) consult stakeholders, primarily through discussion groups. The most consulted actors are government representatives (91%), academic institutions (77%), and healthcare providers (82%). In contrast, patient associations are consulted much less, and the general public even less so (18%, or 4 out of 22 states with consultation processes). Only 26% of states (7 out of 27) have made the conclusions of stakeholder consultations public. This gap in citizen participation raises questions about the social legitimacy and acceptability of deployed AI systems (p. 9-14).
3. Persistent legal and ethical void despite regulatory urgency
Only 7% of EU member states (2 out of 27: France and Italy) have issued specific ethical guidelines for the health sector regarding AI, although 19% (5 out of 27) have cross-sectoral guidelines. 44% of states (12 out of 27) are currently assessing the gaps in their existing laws and policies. Regarding legal liability, only 7% of states (2 out of 27: Spain and Sweden) have issued guidance for manufacturers and users on the application of existing liability regimes to AI. 56% of states (15 out of 27) plan to align their legal liability standards with upcoming European legislation. This legal uncertainty is a major barrier identified by the member states themselves and undermines the trust of professionals and patients (p. 15-24).
4. Inequalities in governance and sharing of health data
67% of EU member states (18 out of 27) have a national strategy on health data, and 63% (17 out of 27) have a national health data hub. However, only 33% of states (9 out of 27) have established rules for cross-border sharing of health data for research purposes. 41% (11 out of 27) have issued guidance on the secondary use of health data for public interest research. The most integrated data sources in national hubs are hospital data (94%), prescription data (82%), and electronic health records (76%), while genomic data is present in only 18% of hubs (3 states: Belgium, Estonia, Latvia). The lack of harmonised frameworks for data sharing limits the development and validation of robust and generalisable AI algorithms (pp. 25-31).
5. Growing but Uneven Adoption of AI Applications in Healthcare
74% of EU member states (20 out of 27) use AI for assisted diagnosis (notably in radiology, dermatology, ophthalmology), making it the most widespread application. 63% (17 out of 27) use conversational platforms (chatbots) for patient assistance. The main motivations are improving patient care (100% of states), reducing pressure on healthcare professionals (96%), and increasing system efficiency (89%). However, only 59% of states (16 out of 27) have identified priority areas for AI, and among these, only 63% (10 out of 16) have allocated specific funding for the development, testing, deployment, and evaluation of these technologies. Adoption is more mature in the EU than in the overall WHO European Region, but remains heterogeneous according to the technical and financial capacities of the states (pp. 31-37).
3. ACTION POINTS FOR LOCAL ACTORS
1. Regularly update national AI strategies to reflect evolving priorities
Local actors must ensure that their national or sectoral AI strategies are periodically reviewed to incorporate technological advancements, sustainability goals, and health system priorities. It is essential to clarify oversight responsibilities by designating entities responsible for implementing the strategies, thereby ensuring accountability, coordination, and continuity across sectors. Policymakers can draw on examples from states that have established formal review mechanisms and clear governance structures (p. 42-43).
2. Strengthen the inclusive and early engagement of all stakeholders, including patients and the general public
To ensure social acceptability and ethical alignment of AI systems, healthcare professionals, patients, developers, and researchers must be involved from the outset of the design and deployment processes. It is recommended to adopt co-regulation models that balance innovation and independent oversight. Local actors can organise public consultations, make the results of consultations transparent, and create formal mechanisms to gather input from patient associations and the general public, who are currently underrepresented (p. 42-43).
3. Develop practical guidance on ethics by design and conduct rigorous impact assessments
States must disseminate practical sectoral guidelines on the integration of ethics from the design of AI systems for health, in order to ensure responsible development, deployment, and alignment with patient safety and their rights. This includes the obligation to conduct data protection impact assessments, ethical assessments, and assessments of fundamental rights. Stakeholders can use available checklists and ethical toolkits and require developers to integrate these considerations throughout the AI lifecycle (p. 43-44, examples: Netherlands, Spain, Italy).
4. Actively prepare for the implementation of the EHDS and harmonise health data governance
By March 2029, all member states will need to implement the governance and technical aspects necessary for the success of the European Health Data Space (EHDS). Local actors must ensure that all stakeholders are timely informed and sufficiently engaged to comply with the obligations arising from the EHDS and ready to explore the opportunities it will offer, particularly for the development and validation of AI solutions in health. This involves creating or strengthening national health data hubs, establishing rules for cross-border sharing and secondary use of data, and harmonising interoperability standards (p. 44, p. 25-31).
5. Establish clear legal accountability frameworks and redress mechanisms for harm caused by AI
States must clearly define the responsibilities of developers, clinicians, data providers, and institutions, relying on mechanisms that allow for swift recourse and accountability when AI systems cause harm. Legal frameworks must cover product liability, personal liability, liability related to input data and data providers, and address causal liability, strict liability, retrospective harm, and vicarious liability mechanisms. Local actors can anticipate European legislative developments and develop rigorous pre- and post-deployment evaluation processes (p. 45, p. 21-23).
4. ADDITIONAL REFERENCES
🔍➕ For more information, see the articles referenced by "Health Practices" on the topic of AI and digital technology ➡️🔗https://pratiquesensante.odoo.com/2-6-intelligence-artificielle-numerique
Ethics and governance of artificial intelligence for health: WHO guidance on large multi-modal models (2024)
Description: Latest WHO guidance (2024) on the ethics and governance of AI for health, specifically on large multi-modal models, complementing the overarching framework from 2021.
5. CROSS-SECTIONAL ANALYSIS — VALUES OF HEALTH PRACTICES
Literacy:The document identifies the need to improve AI literacy through public education programmes, workshops, courses, and partnerships with health agencies (p. 43), but does not propose specific tools tailored to the different levels of understanding of beneficiaries.
Empowerment:Beneficiaries are rarely involved in the design or evaluation of AI systems: only 18% of states consult the general public in their decision-making processes (p. 9-14).
Participation:Co-construction mechanisms mainly exist through discussion groups (81% of states), but they primarily target institutional actors rather than users or communities (p. 9-14).
Community health:The collective dimension is poorly integrated; the report mentions the need to engage the general public and patient associations, but notes their systematic under-representation in consultations (p. 42-43).
Ethics:Cultural and social biases are identified as risks to be addressed through ethical impact assessments, assessments of fundamental rights, and the integration of ethics from the design stage, but only 7% of states have issued specific ethical guidelines for health (p. 15-19).
Human rights:The approach respects the principles of fairness and inclusion through the AI Act, which requires respect for fundamental rights for high-risk systems, and the report recommends impact assessments on fundamental rights (p. 15-24).
Intersectorality:Partnerships are recommended between health authorities, data protection agencies, medical device regulators, and digital health ministries; 48% of EU states have established cross-border partnerships to share knowledge and resources (p. 23-24).
Partnership:44% of states have established rules to facilitate the sharing of health data with private entities for public interest research, but formalised collaboration models remain limited (p. 29-30).
Combating discrimination:The document mentions the need to minimise biases in algorithms through data responsibility practices and high-quality, representative datasets (p. 19-20), but does not detail specific measures against discrimination related to sexual orientation, gender identity, ethnicity, or socio-economic status.
6. EVALUATION OF RESOURCE RELIABILITY
Scientific relevance:Very high. The report is based on a systematic survey conducted in 2024-2025 among 50 member states of the WHO European Region with a response rate of 94%. The methodology is transparent, the data is current (April 2026), and the bibliographic references are recent and verified (all consulted by 10 November 2025). The document is published by the WHO Regional Office for Europe under a Creative Commons licence and co-funded by the European Union.
Operational relevance:Very high. The report provides comparative quantitative data by member state, concrete examples of deployed applications, and action recommendations organised into eight themes directly usable by policymakers and field professionals. The detailed tables (data sources from national hubs, principles covered by ethical guidelines, regulatory approaches) allow for immediate benchmarking and the identification of best practices.
7. MCQ — 5 QUESTIONS
PART 1 — Presentation of the MCQ (without answers)
Question 1:What percentage of EU member states has implemented a national cross-sectoral strategy on AI?
a) 85%
b) 67%
c) 48%
d) 11%
Question 2:What is the major barrier to AI adoption considered by the largest number of EU member states?
a) Environmental impact
b) Legal uncertainty
c) Financial accessibility
d) Data quality and standards
Question 3:What proportion of EU member states with national health data hubs integrates genomic data?
a) 94%
b) 82%
c) 24%
d) 18%
Question 4:How many EU member states have issued specific ethical guidelines for the health sector regarding AI?
a) 15 states (56%)
b) 7 states (26%)
c) 2 states (7%)
d) 0 states
Question 5:By what date must the European Health Data Space (EHDS) be operational for the secondary use of electronic health data?
a) March 2025
b) August 2026
c) March 2029
d) January 2031
PART 2 — Commented correction
Question 1:What percentage of EU member states has implemented a national cross-sectoral strategy on AI?
✅ Correct answer:a) 85%
📝 Explanation:85% of EU member states (23 out of 27) have implemented a national cross-sectoral strategy on AI, while only 11% (3 out of 27) have a sector-specific strategy for health. This cross-sectoral approach is the most widespread but may lack specific focus on health issues. — Source: p. 6-9
Question 2:What is the major barrier to AI adoption considered by the largest number of EU member states?
✅ Correct answer:c) Financial accessibility
📝 Explanation:41% of EU member states (11 out of 27) have identified financial accessibility as a major barrier to the adoption of AI in health. Notably, no member state considered environmental impact as a major barrier, even though AI systems consume a lot of energy and resources. — Source: p. 37-41
Question 3:What proportion of EU member states with national health data hubs integrates genomic data?
✅ Correct answer:d) 18%
📝 Explanation:Among the 17 EU member states with a national health data hub, only 18% (3 states: Belgium, Estonia, Latvia) integrate genomic data. In comparison, 94% integrate hospital data and 82% integrate prescription data, highlighting the lag in integrating genomic data, which is crucial for the development of AI in personalised medicine. — Source: p. 27-28
Question 4:How many EU member states have issued specific ethical guidelines for the health sector regarding AI?
✅ Correct answer:c) 2 states (7%)
📝 Explanation:Only 7% of EU member states (2 out of 27: France and Italy) have issued specific ethical guidelines for the health sector regarding AI. An additional 19% (5 out of 27) have cross-sectoral guidelines applicable to all sectors. This ethical gap is concerning given the stakes of safety and patient rights. — Source: p. 17
Question 5:By what date must the European Health Data Space (EHDS) be operational for the secondary use of electronic health data?
✅ Correct answer:c) March 2029
📝 Explanation:By March 2029, the secondary use of electronic health data must be operational and implemented in all EU member states through the cross-border infrastructure HealthData@EU. This involves the creation of metadata catalogs for a large number of health data categories. This deadline requires member states to actively prepare their infrastructures and governance frameworks. — Source: p. 3, 30-31
8. FREQUENTLY ASKED QUESTIONS (FAQ)
1. What is the European AI Act and how does it apply to the health sector?
The AI Act is the first comprehensive global legislation specifically dedicated to the regulation of artificial intelligence, adopted in June 2024 and coming into effect in August 2024. It establishes a risk-based regulatory framework, categorising AI systems into four levels: unacceptable risk, high risk, limited risk, and minimal risk. In healthcare, most AI applications (diagnostic tools, triage systems, clinical decision support) are classified as high risk due to their potential impact on patient safety and outcomes. From August 2026, and after a transition period of 2 to 3 years depending on the case, high-risk AI systems will have to meet comprehensive obligations including risk management, data quality and governance, traceability, technical documentation, transparency, human oversight, as well as accuracy, robustness, and cybersecurity. — Source: p. 2, 15-24
2. How many EU member states currently use AI for assisted medical diagnosis?
74% of EU member states (20 out of 27) currently use AI for assisted diagnosis, making it the most widespread AI application in healthcare. These systems are primarily deployed in radiology, dermatology, and ophthalmology to enhance imaging and detection. The adoption of diagnostic AI in the EU is more mature than in the wider WHO European Region, reflecting more advanced technical and financial capabilities. — Source: p. 31-37
3. What are the main barriers to the adoption of AI in European healthcare systems?
The three main obstacles identified by EU member states are: financial accessibility (considered a major obstacle by 41% of states, or 11 out of 27), legal uncertainty, and the quality and standards of data. High initial costs (technology, infrastructure, workforce training) pose a particular barrier for healthcare systems with limited resources. Notably, no member state considered the environmental impact of AI to be a major obstacle, despite the high energy and resource consumption of AI systems, particularly large generative models. — Source: p. 37-41
4. What is the European Health Data Space (EHDS) and when will it be operational?
The EHDS is a common framework and data infrastructure supporting data sharing for "primary" uses (healthcare purposes) and "secondary" uses (research, evidence-based policies, regulatory decisions). The EHDS regulation (2025/327) came into effect in March 2025. By March 2029, EU citizens will be able to access and use electronic patient summaries, electronic prescriptions, and electronic dispensations in all member states via the cross-border MyHealth@EU infrastructure. For the secondary use of health data, the HealthData@EU infrastructure must also be operational by March 2029, with the creation of metadata catalogs by all member states for a large number of health data categories (electronic health records, registry data, clinical trial data, reimbursement data). — Source: p. 3, 25-31
5. What proportion of EU member states has established clear legal liability rules for harm caused by AI in health?
Only 7% of EU member states (2 out of 27: Spain and Sweden) have issued guidance for manufacturers and users on the application of existing liability regimes to AI in health. Only 4% (1 state: Belgium) have established a new AI-specific liability regime. 56% of states (15 out of 27) plan to align their legal liability standards with upcoming European legislation. This legal uncertainty poses a major barrier to the adoption of AI and undermines the trust of professionals and patients. — Source: p. 21-22
6. How many member states offer specific training on AI to healthcare professionals?
26% of EU member states (7 out of 27) offer on-the-job training on AI, and 22% (6 out of 27) offer pre-employment training. Only 15% of states provide both on-the-job and pre-employment training. Therefore, training opportunities to enhance healthcare professionals' AI skills remain limited, which hinders the responsible and effective adoption of these technologies. The report recommends integrating AI training into education and professional development, with tiered programmes covering the fundamentals of AI, ethics, data governance, and clinical integration. — Source: p. 9-14, 42-43
7. What are the main sources of data integrated into European national health data hubs?
Among the 17 EU member states with a national health data hub, the most commonly integrated data sources are: hospital data for inpatients (94%, or 16 out of 17), prescription data (82%, or 14 out of 17), electronic health records (76%, or 13 out of 17), administrative data (76%, or 13 out of 17), and mortality data (76%, or 13 out of 17). In contrast, the least integrated data sources are: synthetic data and claims data (24% each, or 4 out of 17), and genomic data (18%, or 3 out of 17: Belgium, Estonia, Latvia). The Czech Republic and Estonia have achieved the highest level of integration with 12 of the 14 commonly used data sources. — Source: p. 27-28
9. REWRITING IN EASY TO READ LANGUAGE
What is this document?
This document talks about artificial intelligence in health.
Artificial intelligence is when a computer helps doctors and caregivers.
The document explains how countries in Europe use this technology.
Why is it important?
Artificial intelligence can help in several ways:
It helps doctors see diseases on X-rays.
It answers patients' questions on the internet.
It reduces the workload of caregivers.
But there are also problems:
It is expensive to buy this technology.
The laws are not clear.
We do not always know who is responsible if the computer makes a mistake.
What does the document say?
About the countries in Europe:
85 out of 100 countries have a plan to use artificial intelligence.
74 out of 100 countries are already using artificial intelligence to assist with diagnosis.
63 out of 100 countries are using talking robots (chatbots) to help patients.
On the rules:
Only 2 out of 27 countries have written ethical rules for health.
Only 2 out of 27 countries have explained who is responsible if artificial intelligence makes a mistake.
On health data:
67 out of 100 countries have a plan to manage health data.
63 out of 100 countries have a special place (hub) to store the data.
Only 33 out of 100 countries have rules for sharing data between countries.
What needs to be done now
For those in charge:
Create clear rules on who is responsible if artificial intelligence makes a mistake.
Train doctors and caregivers to use artificial intelligence.
Ask for patients' opinions before making decisions.
Prepare the system for 2029: by that date, all countries in Europe will need to share health data.
For everyone:
Learn what artificial intelligence is.
Ask questions if you do not understand.
Check that our rights are respected.
Important date to remember
March 2029:all countries in Europe must be ready to share health data with each other.
This will help to better care for patients in all countries.