Artificial intelligence in health systems of the WHO European region: implementation, applications, opportunities, and barriers Authors: Keyrellous Adib, Hanna Emilia Dunning, Nicholas Letchford, Nagui Salama, Yiannos Tolias, Jerome De Barros, Natasha Azzopardi-Muscat, Hans Henri P. Kluge, David Novillo-Ortiz Published: 2026 ABSTRACT Highlights•AI has potential to improve health system quality and efficiency, but adoption remains uneven.•AI-powered diagnostics and conversational tools are the most widely used AI applications.•Legal uncertainty and affordability remain key barriers to wider adoption.•Stakeholder engagement exists but public involvement and training remain limited. Introduction Artificial intelligence (AI) presents significant opportunities to enhance patient care, particularly through early detection, diagnosis, and risk prediction. AI-driven remote patient monitoring allows for continuous tracking of health conditions, enabling timely interventions for individuals with chronic illnesses or those living in underserved areas [1], [2]. For example, AI systems can analyze large datasets to identify high-risk populations and allocate preventive measures more effectively, addressing health disparities. In clinical diagnostics, AI supports physicians by analyzing complex data, such as imaging, pathology, and genetic information, to detect diseases like cancer, cardiovascular conditions, or infectious diseases with improved speed and accuracy. In surgical care, AI-assisted robotics can enhance precision and outcomes [1]. AI is increasingly used to automate time-consuming administrative tasks like documentation, billing, scheduling, and clinical note generation, with language models assisting in drafting medical notes, prescriptions, and radiology reports. These advancements not only improve workflow efficiency but also give healthcare providers more time for meaningful patient interaction [1], [3], [4], [5]. AI empowers patients in managing chronic diseases through tools like virtual health assistants and symptom checkers, promoting better self-care and engagement in their health journey. AI-powered chatbots can offer mental health support, triage symptoms, or guide patients to seek timely medical advice. These applications illustrate the growing role of AI in making healthcare more proactive, personalized, and accessible [1], [6], [7]. Furthermore, AI’s impact extends to pharmaceutical research and drug development. AI systems have accelerated breakthroughs, such as DeepMind’s AlphaFold solving protein folding, which are vital for new therapies and vaccine design [8], [9]. AI in healthcare offers vast opportunities but faces significant barriers [1], [10]. These include a lack of national strategies and policies [1], unclear legal and regulatory frameworks [1], [10], [11], weak infrastructure and data standards [1], [12], limited funding [1], and workforce capacity gaps [10]. Social concerns over job displacement and cultural resistance [1], [10], along with the environmental footprint of large AI models [10], further challenge sustainable adoption. Upskilling the healthcare workforce is central to the WHO European Region’s strategy for harnessing digital health, particularly in the context of AI integration. The WHO Regional Digital Health Action Plan 2023–2030 emphasizes strengthening digital literacy and capacity-building among healthcare professionals to ensure effective use of digital technologies [13]. However, the most recent regional digital health report shows that only 28 Member States offer pre-service digital health education, and only 29 Member States offer in-service training [14]. Moreover, these in-service training opportunities tend to focus mainly on physicians, while nurses, midwives, public health workers, pharmacists, and allied health professionals often receive limited access to relevant digital skills training. Beyond technical proficiency, healthcare workers must develop critical thinking, ethical decision-making, and a clear understanding of AI’s practical applications and risks, such as automation bias [10]. This shift demands a transformation of healthcare workforce education, from rote learning to fostering interdisciplinary competencies, including data governance, AI fundamentals, and communication skills [1], [10]. To support this transition, the development of a new cadre of educators proficient in both health sciences and AI is essential [15], [16]. The objective of this study is to provide an overview of the current landscape, examining how Member States are prioritizing AI in response to health system needs, current barriers and engaging key stakeholders, and addressing critical enablers such as workforce upskilling and preparedness. Methods Study design This study is based on a cross-sectional survey designed to assess the state of AI for health across the WHO European Region. The survey was developed and administered by the WHO Regional Office for Europe and was launched in June 2024, remaining open until March 2025. The survey consisted of 60 items covering key aspects of AI for health, including ethics governance, and implementation practices (see Supplementary 1). The survey instrument combined closed-ended checkbox items with open-text fields to capture contextual details, explanatory information, and relevant references. Data entry was subject to validation procedures, and the survey questions and response options were reviewed by experts from WHO and European Union to ensure alignment with WHO normative guidance and relevant current and forthcoming EU regulations, as well as to reflect the diversity of AI governance and adoption models across Member States. Prior to full survey dissemination, the survey was piloted in two Member States to ensure the validity and feasibility of the instrument, identify potential ambiguities, and confirm that the response options adequately captured the range of national experiences and governance arrangements. The survey was disseminated primarily via a secure digital platform. In instances where direct online submission was not feasible, responses were collected using a structured Microsoft Word document, which was subsequently entered into the digital platform. All survey instructions and questions were provided in both English and Russian. In this survey, AI adoption in health refers to the integration of artificial intelligence technologies into health systems, encompassing their use in the planning, delivery, and governance of health services. Maturity of AI applications in health refers to the stage of development and implementation of AI tools within a health system, typically reflecting the extent to which these tools are operationalized and institutionalized. For the purposes of this survey, the maturity of AI applications in healthcare was categorized as follows:•Informal: limited early-stage adoption or use in a small number of clinical establishments or healthcare settings, typically in the absence of formal governance processes, policies, or national-scale implementation frameworks.•Pilot: time-limited testing and evaluation of AI use cases within a small number of clinical establishments or healthcare settings for specific clinical or operational purposes, including structured pilot implementations beyond purely experimental or research-only activities.•Established: sustained and ongoing operational use in clinical establishments or healthcare settings for a minimum of two years, with the intention or expectation of continued implementation for at least an additional two years. This category includes applications integrated into routine healthcare delivery or organizational workflows. •Informal: limited early-stage adoption or use in a small number of clinical establishments or healthcare settings, typically in the absence of formal governance processes, policies, or national-scale implementation frameworks. Informal: limited early-stage adoption or use in a small number of clinical establishments or healthcare settings, typically in the absence of formal governance processes, policies, or national-scale implementation frameworks. •Pilot: time-limited testing and evaluation of AI use cases within a small number of clinical establishments or healthcare settings for specific clinical or operational purposes, including structured pilot implementations beyond purely experimental or research-only activities. Pilot: time-limited testing and evaluation of AI use cases within a small number of clinical establishments or healthcare settings for specific clinical or operational purposes, including structured pilot implementations beyond purely experimental or research-only activities. •Established: sustained and ongoing operational use in clinical establishments or healthcare settings for a minimum of two years, with the intention or expectation of continued implementation for at least an additional two years. This category includes applications integrated into routine healthcare delivery or organizational workflows. Established: sustained and ongoing operational use in clinical establishments or healthcare settings for a minimum of two years, with the intention or expectation of continued implementation for at least an additional two years. This category includes applications integrated into routine healthcare delivery or organizational workflows. Participants All 53 Member States of the WHO European Region were invited to participate in the survey (Table 1). Each Member State was asked to nominate a national survey coordinator to oversee survey completion. The survey was primarily completed by designated digital health and health information focal points within ministries of health, data authorities, or other relevant governmental agencies. These coordinators were responsible for identifying and engaging relevant national experts in AI for health. The number and expertise of contributors varied by country, and responses reflected collaborative input from multiple stakeholders, rather than an individual perspective.Table 1Member states by subregion (As defined by the United Nations Statistics Division).SubregionMember statesCentral AsiaKazakhstan, Kyrgyz Republic, Tajikistan, Turkmenistan, UzbekistanWestern AsiaArmenia, Azerbaijan, Cyprus, Georgia, Israel, TürkiyeEastern EuropeBelarus, Bulgaria, Czechia, Hungary, Poland, Republic of Moldova, Romania, Russian Federation, Slovak Republic, UkraineNorthern EuropeDenmark, Estonia, Finland, Iceland, Ireland, Latvia, Lithuania, Norway, Sweden, United KingdomSouthern EuropeAlbania, Andorra, Bosnia and Herzegovina, Croatia, Greece, Italy, Malta, Montenegro, North Macedonia, Portugal, San Marino, Serbia, Slovenia, SpainWestern EuropeAustria, Belgium, France, Germany, Luxembourg, Monaco, Netherlands (Kingdom of the), SwitzerlandNote: Only Member States within the WHO European Region are included. For the Western Asia subregion, only those countries that are part of the WHO European Region are considered. Others classified under the WHO Eastern Mediterranean Region are excluded. Bosnia and Herzegovina, Monaco and Turkmenistan did not respond to the survey. Table 1Member states by subregion (As defined by the United Nations Statistics Division).SubregionMember statesCentral AsiaKazakhstan, Kyrgyz Republic, Tajikistan, Turkmenistan, UzbekistanWestern AsiaArmenia, Azerbaijan, Cyprus, Georgia, Israel, TürkiyeEastern EuropeBelarus, Bulgaria, Czechia, Hungary, Poland, Republic of Moldova, Romania, Russian Federation, Slovak Republic, UkraineNorthern EuropeDenmark, Estonia, Finland, Iceland, Ireland, Latvia, Lithuania, Norway, Sweden, United KingdomSouthern EuropeAlbania, Andorra, Bosnia and Herzegovina, Croatia, Greece, Italy, Malta, Montenegro, North Macedonia, Portugal, San Marino, Serbia, Slovenia, SpainWestern EuropeAustria, Belgium, France, Germany, Luxembourg, Monaco, Netherlands (Kingdom of the), SwitzerlandNote: Only Member States within the WHO European Region are included. For the Western Asia subregion, only those countries that are part of the WHO European Region are considered. Others classified under the WHO Eastern Mediterranean Region are excluded. Bosnia and Herzegovina, Monaco and Turkmenistan did not respond to the survey. Member states by subregion (As defined by the United Nations Statistics Division). Note: Only Member States within the WHO European Region are included. For the Western Asia subregion, only those countries that are part of the WHO European Region are considered. Others classified under the WHO Eastern Mediterranean Region are excluded. Bosnia and Herzegovina, Monaco and Turkmenistan did not respond to the survey. Data processing Responses submitted via both the online platform and Microsoft Word were compiled and entered into a unified database by the WHO Regional Office for Europe. All responding Member States that indicated the presence of a national strategy or policy were required to provide supporting documentation, either through the submission of the relevant document or by providing a web link. The dataset was reviewed for consistency and completeness. Where needed, the WHO team contacted national coordinators to clarify unclear or inconsistent entries. Some questions were dependent on member states’ responses to other questions. In these instances, member states that did not respond positively to the initial question would be excluded from the number of respondents on which percentages were computed. Analysis Data were analyzed using an exploratory, descriptive approach, aligned with the primary aim of the study: to describe the strategic and regulatory ecosystem for the use of AI in healthcare across the WHO European Region. The intention was not to compare countries or sub-regions statistically but to understand their current positioning within the evolving AI for health landscape. As such, complex statistical comparisons or inferential methods were not employed. Results are reported as both the percentage of respondents to each specific question and the absolute number of responses received for that question. Data were analyzed at both the regional level and by subregional groupings based on the geographic divisions defined by the United Nations Statistics Division [17]. Additional disaggregation was conducted for the 27 Member States of the European Union (EU27). All analyses were conducted using Microsoft Excel and Python. Results Out of the 53 countries invited, 50 submitted responses, resulting in a response rate of 94%. Opportunities and applications Approximately half of the responding Member States (52%; 26/50) have identified priority areas for the implementation of national AI initiatives, focusing on domains where AI-driven technologies are expected to yield the greatest benefits for health systems and population health. Among these 26 Member States, 54% (14/26) have allocated dedicated funding to support the development, testing, deployment, and evaluation of promising AI technologies. Fig. 1a presents an overview of regional responses showing that improving patient care and health outcomes was the most relevant with 96% (48/50) reported it is of major or moderate relevance. Reducing pressure on the healthcare workforce was the second highest ranked opportunity with 92% (46/50) reported it was of major or moderate relevance. Increasing health system efficiencies was also highly relevant with 90% (45/50) selecting Major or Moderate relevance.Fig. 1Opportunities and Applications of AI for Health. Overview of opportunities and applications of AI in health across the WHO European Region, presented in two panels. (a) The upper chart illustrates the proportion of Member States and their reported levels of relevance regarding the opportunities and motivations driving the development, testing, and use of AI in health. The lower heatmap presents variations in these ratings across subregions. (b) The upper chart shows the proportion of Member States with AI applications in health according to their level of maturity, categorized as informal, pilot, or established. The lower heatmap illustrates variations in the adoption of AI applications across subregions. Opportunities and Applications of AI for Health. Overview of opportunities and applications of AI in health across the WHO European Region, presented in two panels. (a) The upper chart illustrates the proportion of Member States and their reported levels of relevance regarding the opportunities and motivations driving the development, testing, and use of AI in health. The lower heatmap presents variations in these ratings across subregions. (b) The upper chart shows the proportion of Member States with AI applications in health according to their level of maturity, categorized as informal, pilot, or established. The lower heatmap illustrates variations in the adoption of AI applications across subregions. The most common applications of AI in healthcare and their level of developmental maturity are presented in Fig. 1b. Across the WHO European region, the most common AI application was AI-assisted diagnostics (e.g. radiology, dermatology or ophthalmology) used by 64% (32/50), with 30% (15/50) considering it established. An additional 34% (17/50) are currently piloting or informally using AI-assisted diagnostics. The second most common application of AI in healthcare was conversational platforms (“chatbots”) for patient assistance used by 50% (25/50). 24% (12/50) with an established application and an additional 26% (13/50) are currently piloting or informally using AI conversational platforms for patient assistance. As shown in figure 1b, there was a large variation in the adoption and maturity of the AI application across the sub regions. Northern Europe reported the highest levels of implementation across most categories, including AI-assisted diagnostics (100%; 10/10), chatbots for patient assistance (90%; 9/10), and the automation of logistics and administrative tasks (80%; 8/10). Western Europe led in AI-assisted remote patient monitoring, with 57% (4/7) of Member States reporting its use. Barriers to the widespread adoption Fig. 2 below shows an overview of barriers and how Member States across the WHO European Region rated them. Among the barriers, legal uncertainty was identified as the most significant challenge, with 48% (24/50) rating it as majorly important and an additional 38% (19/50) rating it as moderately important. Financial affordability was also an important barrier with 46% (23/50) rating it as a major barrier and an additional 32% (16/50) rating it as moderately important.Fig. 2Barriers to the Widespread Adoption of AI in the Health Sector. Perceived barriers to the widespread adoption of AI in health across the WHO European Regio. The upper chart illustrates the proportion of Member States and their reported levels of importance regarding twelve structural, regulatory, and operational barriers to AI adoption in health systems, categorized as major, moderate, minor, or no importance. The lower table presents variations in the proportion of Member States rating each barrier as of major importance across subregions. Barriers to the Widespread Adoption of AI in the Health Sector. Perceived barriers to the widespread adoption of AI in health across the WHO European Regio. The upper chart illustrates the proportion of Member States and their reported levels of importance regarding twelve structural, regulatory, and operational barriers to AI adoption in health systems, categorized as major, moderate, minor, or no importance. The lower table presents variations in the proportion of Member States rating each barrier as of major importance across subregions. There are notable differences between the sub regions and what they rated as barriers of major importance (Fig. 2). Member States in Western Asia rated data quality and standards as the most important barrier (67%; 4/6). Western Asian Member States also rated the following barriers as majorly important at rates higher than that of other subregions − capacity (50% 3/6), infrastructure (50% 3/6), Trust (50%; 3/6) Job displacement (33%; 2/6) and Environmental impact (33%; 2/6). Stakeholders and collaboration As shown in figure 3, across the region, 72% (36/50) indicated they have engaged with relevant stakeholders on the application of AI-driven technologies in health systems, in one form or another. Focus groups (46%; 23/50) and informal meetings, seminars and workshops (44%; 22/50) represented the most common type of stakeholder engagement. Conversely, public hearings and consultations (16%; 8/50) were the least common approaches to stakeholder engagement. Of the 14 Member States who had not conducted any form of stakeholder engagement, 57% (8/14) were planning to engage stakeholders in the future. Southern Europe has the highest rates of non-engagement with stakeholders (31%; 4/13), however, the majority were planning to do this in the future. Of the 36 Member States reporting at least one form of engagement, a broad range of stakeholders were consulted. The most frequently engaged groups being government actors (81%; 29/36), healthcare providers (75%; 27/36), and AI developers (75%; 27/36), as shown in Fig. 3. Over half of member states, 52% (19/36), most of whom are in northern Europe, consulted more than 6 different types of stakeholder groups. The least common stakeholders to be consulted were patient associations (42%, 15/36) and the broader public (22%; 8/36). Furthermore, 28% (10/36) of Member States have made the insights from their consultations publicly available.Fig. 3Mechanisms of Engagement and Involved Stakeholders in AI for Health. Modes of stakeholder engagement on the use of AI-driven technologies in health systems across the WHO European Region. The upper table illustrates the proportion of Member States and their reported modes of engagement with relevant stakeholders, categorized across approaches ranging from informal workshops and seminars to targeted consultations and formal public hearings. Among Member States reporting at least one mode of engagement, the lower table further disaggregates findings by presenting the types of stakeholders consulted across subregions, encompassing stakeholder groups including health-care providers, regulators, patient associations, and AI developers. Mechanisms of Engagement and Involved Stakeholders in AI for Health. Modes of stakeholder engagement on the use of AI-driven technologies in health systems across the WHO European Region. The upper table illustrates the proportion of Member States and their reported modes of engagement with relevant stakeholders, categorized across approaches ranging from informal workshops and seminars to targeted consultations and formal public hearings. Among Member States reporting at least one mode of engagement, the lower table further disaggregates findings by presenting the types of stakeholders consulted across subregions, encompassing stakeholder groups including health-care providers, regulators, patient associations, and AI developers. Upskilling and AI literacy Fig. 4 shows the proportion of Member States who offer in-service (i.e., after the professionals have been engaged or deployed) and pre-service (i.e., offered in education or training) opportunities to develop AI skills for health and related professionals. Only a quarter of Member States across the WHO European Region (24%; 12/50) have in-service training opportunities for health and related professionals to develop a solid AI skills base. Similarly, pre-service training opportunities are only available in 20% (10/50) across the region. Naturally, there is overlap between the countries offering pre-service and in-service training, and it’s important to note that only 28% (14/50) offer either type of training, including 16% (8/50) that offer both.Fig. 4Health Workforce Education and Capacity Building in Artificial Intelligence. Educational and training opportunities for health and related professionals on AI across the WHO European Region, presented in two panels. The left chart illustrates the proportion of Member States reporting AI-related educational or training opportunities, disaggregated by subregion and categorized according to whether training is embedded within pre-service or in-service requirements. Among Member States reporting in-service training, the right chart further disaggregates findings by presenting the professional groups offered such training, including cadres ranging from doctors and nurses to community health workers and biomedical researchers. Health Workforce Education and Capacity Building in Artificial Intelligence. Educational and training opportunities for health and related professionals on AI across the WHO European Region, presented in two panels. The left chart illustrates the proportion of Member States reporting AI-related educational or training opportunities, disaggregated by subregion and categorized according to whether training is embedded within pre-service or in-service requirements. Among Member States reporting in-service training, the right chart further disaggregates findings by presenting the professional groups offered such training, including cadres ranging from doctors and nurses to community health workers and biomedical researchers. Among the 12 Member States that reported offering in-service training opportunities, Fig. 4 illustrates the proportion of health professional groups that have received such training. All 12 offered in-service training to doctors (100%; 12/12), closely followed by medical information professions (83%, 10/12). A third of Member States (33%, 4/12) have offered training to 6 or more different professions, while 42% (5/12) have offered training to 4 or less professions. Furthermore, across the European Region, 42% of Member States (21/50) reported they have created new professional roles and opportunities for people with in-depth knowledge of data science and AI in the health sector. Discussion Opportunities and applications The study highlights the alignment between current applications of AI in national health systems and the immediate priorities − improving patient care and health outcomes and reducing pressure on the healthcare workforce. With nearly two-thirds of countries employing AI-assisted diagnostics and screening, the implications for healthcare professionals are promising. Evidence indicates that AI integration in screening processes can reduce radiologists’ workloads by approximately 50%, without compromising detection performance or diagnostic accuracy [18], [19]. A scoping review of AI-assisted diagnostics found improved detection rates and accuracy [20], but few studies assessed direct patient outcomes, and evidence linking diagnostic gains to long-term health benefits remains limited, while risks include biased or low-quality outputs, automation bias, erosion of clinical skills, and challenges in obtaining informed consent [10], [20]. AI-powered chatbots are increasingly used to support patient self-management by providing medical information, personalized guidance, symptom triage, mental health support, and clinical trial matching [21]. While they have potential to enhance health literacy and autonomy, they may affect the clinician–patient relationship by reducing meaningful engagement and professional oversight [10], [22], [23]. Additional risks include misinformation, biased or incomplete outputs, exploitation of user vulnerabilities, epistemic injustice affecting marginalized groups, and language limitations that increase miscommunication for non-English speakers [10], [24], [25], [26], [27], [28]. Barriers to widespread adoption This study found that the greatest barrier impeding the widespread adoption of AI in the health sector is legal uncertainty. The legal uncertainty surrounding AI in European healthcare arises from several reasons including the misalignment between rapidly evolving technologies and the slower adaptation of existing legal frameworks [29]. This contributes to significant ambiguity around liability, particularly concerning responsibility for errors made by AI systems, whether it lies with the developer, clinician, or provider institution [30]. Further complications arise as AI’s data-intensive nature clashes with the EU’s core GDPR principles like data minimization and informed consent [31]. An additional layer of complexity arises from the growing accessibility of AI technologies in healthcare beyond formal health systems, including through public platforms and private enterprises. This expansion blurs the boundaries of existing regulatory categories, positioning such tools within a regulatory grey area between rigorously controlled clinical applications and more loosely governed wellness products [10]. This study reinforced that affordability remains a critical barrier to scaling AI in healthcare. High upfront costs for infrastructure (e.g. secure data systems, cloud computing, and connectivity) can limit the adoption of AI systems and tools [32]. Many health systems also lack sustainable funding models and reimbursement mechanisms to support long-term AI integration [33] Workforce training adds another layer of expense, as upskilling clinicians in digital competencies and AI literacy requires ongoing investment [34]. Moreover, the limited availability of cost-effectiveness evidence for AI tools makes it difficult for policymakers to justify financial commitments [35]. Besides, Access to large multi-models (LMMs) may be restricted by their cost, as many are available only through paid subscriptions due to the high expenses associated with their development and operation [10]. Engagement and collaboration Effective collaboration between governments, health professionals, and AI developers is essential to ensure that AI in health care is safe, ethical, and aligned with public needs [1]. In particular, the involvement of health professionals helps ground AI development in clinical relevance and public health priorities [36], [37]. As AI technologies grow more complex, particularly with black-box models, co-regulation models have emerged where governments and companies share oversight. While this can improve understanding, it may also compromise independent regulation and lead to over-reliance on corporate self-regulation [1]. Similarly, Public-private partnerships (PPPs) offer a way to harness private innovation while maintaining government oversight [1]. However, they face challenges such as information asymmetry, conflicting objectives, and concerns over data use and fair benefit-sharing [1]. Our findings highlight a significant gap, frequently noted in the literature, concerning the limited inclusion of patients and public associations in current discussions on AI in healthcare [38]. The lack of meaningful engagement throughout the AI life cycle, from development to deployment, can undermine the relevance, acceptance, and trust in AI-driven solutions. Without grounding these tools in real-world needs, there's a greater risk they will lack practical or clinical value [38]. Limited involvement of patients and end users often results in poor buy-in, resistance to adoption, and weak integration into care settings. Overlooking these perspectives increases the chance of creating technically advanced tools that fail to meet the expectations and needs of those they aim to serve [38]. Upskilling and AI literacy This study found that less than a quarter of countries in the WHO European Region offer pre-service or in-service training opportunities for health and related professionals to develop a solid AI skills base. A meta-analysis study highlighted the Primary care providers demonstrate generally low knowledge of AI-assisted digital health tools, with learning often occurring passively and accompanied by misconceptions [39]. Despite this, approximately half express a willingness to adopt such tools in clinical practice, although concerns about job security, infrastructure, and ethical implications persist. Without targeted, practice-oriented training, clinicians may struggle to understand AI’s role, capabilities, and limitations, increasing the risk of misuse or overreliance on AI tools [40]. Russell et al. [41] identified six core competency domains for health professionals using AI in healthcare, covering AI knowledge, ethical and social implications, patient-centered clinical use, tool evaluation, workflow adaptation, and continuous learning for practice improvement. In the absence of such competencies, healthcare professionals face the risk of automation bias, potentially overlooking errors or delegating complex moral decisions to AI systems [1], [10]. Over time, the overreliance on AI tools may lead to skills degradation, eroding clinicians’ confidence and ability to critically assess AI outputs, intervene when necessary, or manage care independently during system failures [1], [10]. Therefore, structured, competency-based AI training is not only necessary to bridge current gaps but also vital to safeguard clinical judgment and professional integrity in an AI-enhanced healthcare environment. Strength and limitations This is the first study to apply a systematic, national-level approach to capture a regional overview of AI adoption in health across the WHO European Region, including: uptake, barriers to adoption, engagement modalities, stakeholder involvement, and health workforce preparedness. Nonetheless, several limitations should be acknowledged. A primary limitation stems from the variability of definitions used within the survey questions, which may have been interpreted differently by each participating Member State. Moreover, the validity of the survey findings depends significantly on the expertise of the designated national coordinators and the consulted experts across various AI-related domains. To mitigate this, respondents were encouraged to provide supporting documentation and relevant links where available. A clear definition of AI maturity levels was also included to support more consistent national-level ratings. Another limitation is the inherently subjective nature of ratings concerning key opportunities and major barriers, which reflect the perspectives and knowledge of individual respondents. However, the collaborative nature of the survey process, where survey coordinators convened multiple agencies and stakeholders to formulate a consolidated response, likely contributed to enhancing the validity and reliability of the data collected. Policy implications AI applications in health should be designed with clear, reliable, and context-appropriate functions that align with patient interests and national health goals, with transparent communication of their capabilities and limitations. Strengthening funding mechanisms, creating implementation roadmaps, and integrating AI tools into existing health system workflows are essential to maximize impact. Establishing accreditation and certification standards for developers, alongside independent pre- and post-deployment impact assessments and ongoing surveillance, ensures quality, detects bias or performance drift, and mitigates potential harms, with results made publicly available to enhance accountability and trust. Regulatory sandboxes can provide controlled environments to safely test AI applications before widespread deployment, while thorough assessments of cost-effectiveness and alignment with ethical and human rights standards ensure that AI solutions deliver value and uphold fundamental principles prior to adoption. Equitable and accountable public–private partnerships should operate transparently, disclose agreements publicly, uphold individual and community rights by securing ownership or access to AI technologies, and clearly define which healthcare responsibilities remain public versus delegated to private actors. Inclusive and transparent engagement should be embedded throughout AI development by involving end users, the public, and industry in co-design and co-regulation processes to surface ethical concerns, enhance accountability, and build trust. Creating platforms and dialogues that improve transparency around data sharing and promote culturally acceptable AI applications can further foster a shared understanding of AI and its use of data. AI literacy should be institutionalized by integrating AI-related content into pre-service curricula, in-service training, and continuing professional development to equip the health workforce with a solid understanding of its benefits, risks, and ethical considerations. Continuous education for relevant stakeholders is essential to ensure ongoing compliance with evolving ethical, legal, and regulatory requirements, embedding these considerations throughout the AI design lifecycle. Disclaimer The information and views in this article are those of the authors and do not necessarily reflect the official opinion of the European Commission or WHO. CRediT authorship contribution statement Keyrellous Adib: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Software, Project administration, Methodology, Formal analysis, Data curation, Conceptualization. Hanna Emilia Dunning: Writing – review & editing, Writing – original draft, Visualization, Validation, Formal analysis, Data curation. Nicholas Letchford: Writing – review & editing, Writing – original draft, Visualization, Methodology, Formal analysis, Data curation. Nagui Salama: Writing – review & editing, Visualization, Validation, Formal analysis, Data curation. Yiannos Tolias: Writing – review & editing, Investigation. Jerome De Barros: Writing – review & editing, Investigation. Natasha Azzopardi-Muscat: Writing – review & editing, Investigation. Hans Henri P. Kluge: Writing – review & editing, Investigation. David Novillo-Ortiz: Writing – review & editing, Validation, Supervision, Resources, Project administration, Investigation, Funding acquisition, Conceptualization. Funding We would like to thank the European Commission for supporting this work under the project “Supporting Member States in strengthening health information systems and boosting health data governance”. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.