Scaling Artificial Intelligence in Health

2. AI in Health Policy Checklist

Adopting AI solutions in healthcare needs pace, focus, and buy-in to ensure that health systems can scale fast while doing no harm. This approach must create space for vision, alignment, multi-stakeholder collaboration and collective impact. Health is a fundamental pillar of society, yet in this domain, harm can be irreversible – even fatal – raising the stakes for responsible adoption of AI.

Much of the challenge in health stems from needing to balance several forces at the same time – advancing science and innovation, doing no harm, and providing equitable outcomes for all. As discussed in Chapter 1, the use of AI in Health simultaneously comes with significant opportunities and significant risks. Many health systems – and individual institutions – have tried to find this balance point in isolated siloes. While delivering against local objectives, this has the potential to harm advancement for the health system as a whole – delivering sub-optimal outcomes for individuals and populations.

Other industries have learned that collaboration on data infrastructure and digital foundations not only reduces operational costs – it also enables targeted investments in areas that offer long-term strategic value and improve customer experience. For example, co‑operation in the financial sector enables individuals to withdraw cash from bank machines and move funds around the world. Similarly, co‑operation among airlines simplifies the ability for consumers to book flights themselves and for airplanes to fly safely across the globe. These industries have collaborated to align on data and technical foundations while competing on design and implementation of advanced practices. This led to improved services, lower overall cost, and innovation that thrives.

In health, the lack of alignment on policy, data, and technical foundations prevents health institutions from collaborating easily. Fragmentation in foundations acts as a barrier to innovation and inhibits the ability of health systems to move toward person-centric health systems that improve outcomes. A lack of effective AI governance and cross-border compatibility in policy inhibits the potential of AI to scale in the digitalised health ecosystem. This increases costs and degrades outcomes. This checklist seeks to help policymakers by identifying potential blind spots so policy action can support AI solutions that could scale efficiently and enable better health outcomes for all.

Advancing an AI in Health Policy Checklist

As AI becomes increasingly embedded in health systems, there is growing recognition of the need for a co‑operative approach to guide responsible scale of AI in health. While general AI principles offer valuable high-level direction, the complexity of health systems – which seek to meet the quintuple aim for healthcare while respecting the sensitivity of health data – demand guidance rooted in real world implementation needs (Nundy, Cooper and Mate, 2022[1]). The AI in Health Policy Checklist provides guidance on a co‑operative approach.

The purpose of the AI in Health Policy Checklist is to identify the foundational policy areas that support the responsible scale of AI and to understand countries’ readiness in their adoption of AI. This can help policymakers deliver a consistent and sustainable approach. Countries can also work together to identify how to address the compatibility of policy areas across borders. That compatibility will accelerate the scale of innovation to optimise both human and economic benefits. This checklist does not distinguish between specific AI modalities. Rather it provides guidance for ensuring that AI integration into health systems is ethically grounded, operationally efficient and clinically effective. Future work for this checklist could examine challenges with specific modalities (such as Agentic AI) or applications (such as in public health).

The checklist is anchored in cross-border collaboration and practical relevance. The process of development drew on desk research, surveys, international partnerships and iterative consultation to identify key policy areas. Key policy areas were identified to support the responsible scale of AI in health across core policy areas. To translate these policy areas into actionable elements, the process drew on the OECD’s Integrated Digital Health Ecosystem Policy Checklist (OECD, 2023[2]) and the OECD’s Framework for Trustworthy AI in Government (OECD, 2024[3]). These tools provided a foundation for categorising key concepts under the thematic pillars of enablers, guardrails, engagement, and trustworthiness (See Figure 2.1).

Figure 2.1. Policy checklist to support responsible scale and scale of AI in health

Four thematic pillars of an AI in Health Policy Checklist

An AI in Health Policy Checklist emphasises the following areas, building on OECD Governing with Artificial Intelligence (OECD, 2024[3]).

Enablers for AI in health develop strong data and digital foundations to support timely and interoperable health data across the health sector and government services. Data are then accessible for use in the development, testing, implementation, and evolution of AI solutions while ensuring continuous safety of those solutions. Processes are streamlined to review, validate, approve, deploy, and evolve AI in health solutions, with sufficient and sustainable human and technical capacity.

Guardrails support the safe and responsible deployment of AI solutions. The guardrails utilise agreed upon common objectives for an AI in Health programme across all stakeholders and allow all parts of the health system to work together toward agreed goals. The guardrails also establish governance structures that put incentives in place to deliver those objectives. Procedures are put in place to measure, monitor, and address issues with the effectiveness of the AI in Health program. Ultimately, this establishes clear responsibilities across the AI lifecycle, including who is accountable for doing what with which controls, and establishing the implications for issues as they arise in the interest of enabling efficient and effective value of AI while protecting from harm.

Engagement for AI in health relates to the involvement of key stakeholders in the responsible scale of AI. Key stakeholders include the public, healthcare providers, and other enabling partners. Engagement involves building the capacity and capability to engage with the AI in Health program. A key outcome of engagement is enabling trustworthiness to accelerate the adoption of approved AI solutions.

In addition, consideration of Trustworthiness for AI in health is necessary for consideration of bioethics and responsible scale. A fulsome approach to trustworthiness and ethics will consider action from multiple perspectives – the potential harm from both the implementation and the delay in implementation on AI in Health solutions. Embedding ethics would reflect the OECD’s AI principles (OECD, 2024[4]) and the broader perspectives of bioethics reflecting patient beneficence, patient non-maleficence, patient autonomy, and justice.

With this as background, the following section describes each of the parts of the OECD AI in Health Policy Checklist. This is based on the nine areas noted in Figure 2.1. With the description of the policy areas there is a brief discussion of why it is important; reflection on emergent, or divergent practices; and notable examples of the policy areas in practice. A table with questions for policymakers to understand their readiness for the responsible scale of AI is included at the end of this chapter. Chapter 3 builds on the policy checklist by examining progress across OECD countries for the responsible scale of AI.

AI in Health Policy Checklist

Enabler: Use and protection of health data in AI solutions

Why this is important: Data underpins all stages in the design, development, implementation, use, evaluation, and evolution of AI solutions. Health data represents the largest segment of data from any industry (estimated at 30%) and is growing faster than any other sector (Weber, 2024[5]). Yet, the health sector is using less than 5% of the data for decision making purposes (World Bank Group, 2023[6]). Health data has the unique property of simultaneously providing value for individual care (such as through direct care) and for communities and the population (such as through public health measures, research, innovation, or health system monitoring). It is important for enablers of health data to account for the risks associated with both data use (e.g. privacy breaches) and data non-use (e.g. preventing safety issues, supporting innovation).

Further, many digital health investments have faced issues achieving effective return on investment in part due to the significant fragmentation of data and digital assets within and across health systems. This fragmentation has made it challenging to have comprehensive trusted information about individuals for the provision of care. Fragmentation also complicates data compilation and aggregation – with appropriate protections – for health system safety and performance, public health, research, and innovation among others. These challenges slow effective health interventions, prevent scientific advancement, and increase costs due to unnecessary investments in data collection, storage, and integration. To realise the benefits of AI, it is critical to strengthen data and digital foundations so that data are representative of populations and the capacity exists to use AI solutions at scale.

What this includes: There are many considerations for data in healthcare, including that it is findable, accessible, interoperable, and reusable (FAIR) (GO FAIR, 2021[7]). Enablers for the use of health data could be operationalised for data to be:

• Findable: Data assets intended for analytic use could be catalogued and easily discoverable by authorised users. Catalogues should be annotated to support the search of data assets for re‑use. Improving the findability of high quality, representative datasets will support AI developers, researchers, and policymakers to access existing resources, reduce redundant investments, and speed up innovation. This is especially important in the current landscape of international and cross-border collaborations, in areas of public health, research and healthcare innovations.

• Accessible with individual consent: Establish clear practices for individuals to consent for the use of their health data in care and analytics. Consider the appropriate use of opt-in or opt-out consent, or the use of compliant alternatives when consent is impractical while achieving health system objectives (such as individual emergency care or emergency public health management).

• Accessible health data for primary use: Health data should be accessible to authorised healthcare providers, individuals, and/or their delegates for direct care, aligned with individual consent and considerations for privacy and safety. How data are held and governed should also be addressed while shifting culture from risk avoidance and protection to data stewardship, which prioritises optimising the collection, use, and re‑use of data in individual and public interest while providing appropriate protections across data supply chains. This process supports integrated care, professional collaboration, and patient trust (Digital Transformations for Health Lab, 2025[8]).

• Accessible health data for analytic use in the public interest (e.g. public safety, health system improvement): Data intended for analytic uses can be accessible through clear, streamlined processes that minimise administrative barriers while protecting data privacy and security. Facilitating streamlined access across health institutions supports the responsible use of data in the public interest in research, public health initiatives, and innovation. These could be governed by robust data protection and accountability frameworks that support trusted cross-border collaboration. A key tool in this policy area is consideration for privacy enhancing technologies (PETs) (e.g. pseudonymisation, encryption) (OECD, 2025[9]; OECD, 2023[10]). Decisions regarding data access and analytic use should follow transparent, accountable processes, with clearly defined criteria – supported by mechanisms that demonstrate how confidentiality, intellectual property and commercially sensitive information will be protected throughout the data lifecycle across the public and private sectors.

• Interoperable: Health systems should be interoperable. This does not mean that every country adopts the same standards for data capture and exchange, rather, it means that the standards that are adopted are compatible with other standards in the areas that are most important for the provision of care and crucial analytic uses such as research, clinical trials, or public health. International co‑operation between standards bodies, governments, and cross sector stakeholders is beneficial to enabling interoperability. Governments could support this by adopting globally recognised standards, along with technologies that enable data exchange across diverse systems while accommodating local needs.

• Data quality: Health system data should have its quality measured and reported. Given that data are the foundation of AI, if data are of poor quality, then the resulting AI solutions will be of poor quality. Measuring and reporting data quality will help international collaborators work together with confidence while building trust with stakeholders. This quality evaluation would measure areas of skewness, representativeness, accuracy, traceability, and timeliness. Consideration could also be given to whether data quality measures are globally aligned to support international collaboration, while still allowing for local components that reflect specific population needs.

• Linkable: National health systems can adopt a consistent privacy preserving digital identifier to securely link data about individuals regardless of the location or mode of their care, while safeguarding their privacy. This will help people who receive care across borders to increase trust and ensure completeness of their health records. Linking data is also critical for AI development, where linking data across domains strengthens the quality, representativeness, and trustworthiness of AI solutions (OECD, 2025[11]).

• Sovereign and Secure Storage: National health systems can clarify expectations for the security and storage of data, including how data are encrypted – both in motion and at rest – and where data are stored to protect data sovereignty. This helps protect sovereign interests while supporting international collaboration in critical areas such as digital security.

Emerging leading practices: The enablers above fall under the umbrella of health data governance, though there is variation in how these are implemented. The European Health Data Space (EHDS) and the identification of country-led health data authorities is a strong step forward. As countries work together to align their national practices with the EHDS regulation, they will help to define the leading practices. Some emerging leading practices such as the adoption of the International Patient Summary (IPS) for patient records demonstrates how international collaboration could operate with a foundation of open standards and interoperability (The International Patient Summary, 2025[12]). In addition, the Digital Health Model Security Notice developed by the Global Digital Health Partnership (GDHP) provides consistent requirements to vendors of digital security expectations (Global Digital Health Partnership, 2023[13]).

Enabler: Define and operationalise the assurance and use of AI in health services

Why this is important: The OECD AI Principles (OECD, 2024[4]) provide clarity on what is expected for the deployment and use of AI across the entire AI lifecycle. Specifically, that AI solutions should advance Inclusive Growth, Human Rights, Transparency and Explainability, Security and Safety, and Accountability. While these terms are often used to describe trustworthy AI, incompatibilities in the interpretation of these terms can prevent the scalability of AI solutions when they need to be adjusted to meet different requirements.

For innovative AI solutions in health to scale across broader populations, co‑ordination of assurance and approval processes is essential. This can be supported through cross-country agreements on shared principles, compatible assurance standards, and transparent evaluation procedures. Such co‑ordination does not imply that approval in one country automatically results in approval in others, but rather reflects the principle of “reliance”, where assurance processes are designed to be compatible. This approach enables AI innovators to understand the steps required to gain approval, facilitating the scaling of solutions across borders. It also helps reduce duplication, accelerate access, and foster economic growth.

What this includes: Enablers for the AI life cycle can be clarified to enable:

• AI solution risk management: A key component of AI assurance processes includes the assessment of the risk of the AI solutions in the context of the health system. Questions such as “what is the likelihood of a negative occurrence?”, “what is the likelihood of not achieving a positive occurrence if we do not act?”, “what is the impact of negative or positive occurrences?”, and “what actions are in place that lower negative likelihoods, increase positive likelihoods, and minimise negative impacts” are all considerations. Compatibility for risk management – and the identification of appropriate actions given the risk level – help innovators understand what controls need to be incorporated into their solutions to provide greater certainty in their design. This also helps health stakeholders build trust in the use of AI solutions through the implementation of controls.

• AI solution assessment: Health systems can clarify the processes for the assessment of AI solutions – establishing requirements for AI solutions to be considered, reviewed, and accepted as an approved solution. The assessment should include the classification of what constitutes AI as a medical device in health vs. AI as a non-medical device in health and criteria for their review and approval. This assessment also includes the intended use for the AI solution in terms of its geography, target population, and other factors. This could include direct-to-consumer vs. direct-to-provider vs. administrative solutions. Review would assess the cost-effectiveness and other implications of AI solutions including through market-accepted tools such as Health Technology Assessments (HTA). Health systems could also align on tool effectiveness, value assessment and reimbursement, and ongoing compliance frameworks to promote the long-term viability of integrating AI into clinical practice and encourage investment in scalable, sustainable innovation. Given that AI solutions are frequently updated, based on training data, the criteria for periodic updates to and review of the AI solutions could be defined.

• Algorithmic useability: Assure that AI solutions are fair in their application, ideally with reduced skewness, or at minimum understood representativeness of the data used to develop the solution and its alignment with anticipated used of the AI solution. Further, AI solutions should provide sufficiently informative output to help users understand the AI decisions depending on the context of the AI solutions and individuals either using or impacted by the solution to facilitate their use of results from the AI solution (e.g. explainable and transparent).This facilitates having a human-in-the‑loop during the use of AI in health.

• AI Model cards: Health systems could have methods to clearly communicate the result of the assessment to inform consumers prior to deciding to use the AI solutions. A model card, a standardised document summarising key information about an AI solution, including its intended use, performance, limitations and risks, can help achieve this. This would include details on skewness, performance metrics and representativeness, as well as transparent disclosure of any residual skewness, the reasons it persists, and its acceptability for specific subpopulations. Given the adaptive nature of AI systems, model cards should be treated as living documents, updated regularly to reflect model changes, retraining and real-world performance over time. International collaboration is useful to articulate their trustworthiness for the consumer and health professionals alike.

• Procurement: Health system procurement could be modernised to leverage established networks to simplify procurement for individual health facilities (e.g. through certification models). New procurements could be updated to articulate the requirements of publicly funded health systems as new technologies, such as AI, are introduced into their footprint. This could include areas noted in the previous sections around data management and security (including interoperability), adoption of model cards, algorithmic viability, and assurance models. A modern approach to procurement helps amplify a consistent signal to the vendor community of expectations and requirements.

Alongside assurance and procurement, additional considerations are:

• Patient consent for AI in clinical use: Define when consent for the use of AI is necessary (vs. appropriate alternatives) and how that consent is received, recorded, and shared.

• Approval for AI in clinical use: Guidelines can be provided to health systems with recommendations on when AI solutions could be used due to the availability of evidence, safety requirements, and appropriate evaluations. These recommendations would be contingent on contextual factors such as patient characteristics and their point along the care pathway. This guidance aims to clarify when human oversight is required and when AI solutions may be autonomous with appropriate review.

• Reimbursement of AI in clinical use: Clarify when the use of such solutions require special re‑imbursement (as opposed to being a tool in the delivery or care), (Lobig et al., 2023[14]), (van Kessel et al., 2023[15]).

• Liability for AI in clinical use: Define how liability is managed, such as how to address a negative situation that arises from the use of the AI solution. Could be extended to consider both direct-to-clinician and direct-to-consumer models.

To accelerate innovation, many countries are establishing approaches for:

• Regulatory sandboxes: Regulatory sandboxes test innovative solutions that are on the “edge” of current practices. The sandbox provides extraordinary oversight to prevent harms, while establishing a “proving ground” for what is innovation possible with regulatory change or clearer interpretation. Implementation of a regulatory sandbox includes adjusting approval processes post-test to avoid lengthy delays for critical technologies while sufficient evidence is gathered.

Emerging leading practices: A leading practice is the adoption of model cards – via Conformité Européenne (CE) marking in Europe, which certifies compliance with the European Union (EU) safety, health, and environmental protection standards (European Commission, 2025[16]) and the Coalition for Health AI (CHAI) in the United States which has introduced model card frameworks to enhance transparency and accountability in health-related AI systems (CHAI, 2025[17]). The International Medical Devices Regulatory Forum (IMDRF) is currently in the process of designing a comprehensive guide to AI-regulation across the AI lifecycle (IMDRF, 2024[18]).

Areas where more convergence is needed: Countries are developing idiosyncratic approaches to risk management and the definition of medical devices. For example, in the United States, the FDA has adopted risk management practices based on the output of the AI solution whereas the European Union (EU) considers both the input and output of AI solutions. This divergence has led to cases where AI tools like the AI Scribes were originally considered high risk in the EU (due to using personal health information for training) and low risk in the United States (due to doctors being in the loop). These differences highlight a key challenge for international collaboration, that without the shared understanding of risk categories policy coherence and scalability of AI solutions remains limited. Establishing agreement on such risk categories is a necessary pre‑condition for developing interoperable AI risk governance and scalability of AI solutions, perhaps leveraging the risk management framework for medical devices published by the International Medical Devices Regulator Forum (IMDRF) (IMDRF, 2025[19]; Chapman, 2025[20]).

Enabler: Sufficient human and technical capacity and capability

Why this is important: Given the wide range of policy and technical activities that are needed to help the design, implementation, operation, monitoring, and improvement of AI in health solutions – and the significant opportunity from the responsible scale of AI – there is a need to build human and technical capacity. Human capacity and capability are important across borders to support the transferability and mobility of skills. Technical capacity is beneficial to share sufficient and available computing power.

What this includes: To build sufficient and capable human and technical capacity, countries may establish practices for:

• Planning for future workforce: Consider the needs of the health workforce as part of the integration of AI: what are the impacts on the frontline? What are the impacts on patients? What supporting functions are necessary? and What oversight functions are required for the system to be safe? There is benefit to proactively planning and acting on future health workforce needs, for example considering future hybrid roles. Planning could assess current gaps in tools and knowledge across different workforce functions. Targeted upskilling programmes can be developed to minimise knowledge gaps and prepare the digitally enabled health workforce.

• Planning for public knowledge needs and trust: Consider the needs for the public as AI is introduced into the health system. Focusses on building sufficient knowledge to enable the public’s capacity to effectively engage with AI. This planning process recognises that trust and understanding are mutually reinforcing. Planning could assess the current gaps in tools and knowledge across the public and include developing methodologies, approaches and training programmes to prepare the public to be empowered participants in their care with AI.

• Planning for sufficient technical capacity: The development of AI solutions requires sufficient computing power and storage to operate effectively. Decisions would need to be made about whether to insist on local computing capability or whether co‑operation is better to share computing capacity with partners, for example with cloud computing and enabling federated learning. Planning could assess current gaps in technical capacity, project future demand, and develop forward-looking strategies for building technical capacity when and where it is needed.

Emerging leading practice: Few countries in the OECD actively project and plan for the health workforce that is required for the integration of AI into health. In that work, new planning could include technologists, data managers and stewards, privacy officers, policymakers, as well as administrative staff and office managers. In the United Kingdom, the Digital and Data Profession Capability Framework has been developed to build capacity in support functions, though is not specific to the health sector (GOV.UK, 2025[21]). By doing workforce planning co‑operatively the definition of new roles and evolution of existing roles will be more consistent which will simplify accreditation and transferability of health workers.

Guardrail: Operationalise action toward common vision and objectives

Why this is important: Health systems benefit from having an agreed common objective for the design of their AI in health ecosystem that is agreed across all stakeholder groups. An agreed common objective helps decisions about priority and action to evaluate those decisions against their contribution to the agreed common objective. Across countries, specific agreed objectives will vary; however, they will likely adopt common attributes in being person-centred and rights-based (to health, to privacy, and to benefit from science).

What this includes: Responsible scale of AI into health systems benefit from common objectives defined by:

• Establishing a vision: Governments could clearly articulate the intended outcome from investments in AI for health. This could consider impacts on people, providers, and industry – and the role of government to guide actions to achieve the vision. Based on the OECD Health Minister Declaration (2024), health systems should be centred on people and provide personalised care (OECD/LEGAL/0500, 2024[22]; WHO, 2021[23]). Health systems would preserve and enhance human-decision making while promoting human and societal well-being.

• Establishing a strategy or action plan: As will be seen in Chapter 3, countries are starting to develop specific strategies or action plans at the intersection of AI and health to deliver against agreed objectives for AI in health. Such strategies direct collective actions toward balancing the forces of market, health, and the public.

Many of the strategies or action plans adopt common ways of working that could be considered:

• Open: Open approaches include where data, digital, and AI infrastructure are as open as possible and closed as necessary. This approach fosters protection and the use of health data resources for systems outcomes. This will help systems have fair access to data while minimising skewness and discrimination. Humans benefit from AI solutions that are trained on representative data sets.

• Collaborative: Stakeholders co‑operate to share knowledge to achieve better health outcomes (WHO, 2021[23]). Systems also work together toward the responsible scale of trustworthy AI (OECD, 2024[4]). In health, this helps private and public sector organisations have collaboration mechanisms that are inclusive, transparent, and open and result in better compatibility of approaches within and across borders. This collaboration will encourage the involvement of stakeholders, including patients, the public, and underserved communities, and foster transparent and understandable processes for AI solutions to effectively scale and be responsibly adopted.

• Responsive and adaptable: Policies and processes are designed to be responsive and adaptable to new opportunities and risks as they arise. Speed of change is calibrated to the risk of action and risk of inaction. Given the increasing pace of innovation, this may require forms of adaptive governance such as rules being set dynamically by designated authorities with appropriate transparency, communication, and oversight rather through specific regulation or legislation. Responsiveness also understands the range of types of AI solutions (e.g. LLMs, agentic AI, predictive analysis, etc.) and using a risk-based evaluation, helps to prioritise action to respond to the size of the risk or opportunity.

Emerging leading practice: As will be discussed in Chapter 3, an emerging leading practice is for health systems to develop a strategy or action plan at the intersection of AI and health to ensure progress that is in alignment with common health system objectives.

Guardrail: Oversight, measurement, and monitoring of AI solutions

Why this is important: A key method to assure the ongoing safety and effectiveness of AI in health solutions is to measure their performance in clinical settings and to monitor those metrics to act when certain thresholds are attained. When measurement approaches are aligned across jurisdictions, multi-country assessments and co‑operation become more feasible. More broadly, consistent measurement – alongside open sharing of results – simplifies post-market assessment, supporting collaborative learning and improvement. This would be grounded in governance structures supported by empowered institutions. Performance measurement also supports the tracking of benefits from the use of AI in health and progress against strategic milestones.

What this includes: Measurement and monitoring of AI solutions would include:

• Evaluating post-market effectiveness: In the approval for the AI solution, there is generally an effectiveness measure that describes the expected clinical performance and measures to verify the stability of the AI solution both before and after deployment. Post-market effectiveness tracks the AI solution for performance drift and establishes pathways for notification and remediation. Measurement plans and reporting cycles post-deployment may include safety plans (reviewing for unintended and harmful outcomes), patient-reported experience and outcomes and socio-demographic uptake and impacts. That is, measuring for the availability, accessibility, use, and impact of the AI solution by populations. Serious incidents could be transparently reported and acted on to prevent future re‑occurrence.

• Benefits assessment: This would measure and estimate the net benefit from the AI solution. This would be related to the benefits hypothesis and/or quantified targets developed at the approval of the AI solution. Benefits could span the quintuple aim, including aspects of outcome improvement, equity uplift, or system efficiency. Benefits would be related to system planning and re‑investment, including benefits from the scale of approved AI solutions. For example, this could estimate changes in workforce productivity and help understand how that productivity gain is re‑invested – whether in reducing burnout, investing more time in care, and / or in seeing additional patients.

• Energy usage: This would measure the energy impact and usage of AI in health solutions to understand the environmental impact of solutions from energy used in the end-to‑end life cycle of AI processing. This includes evaluating future energy demands during development, deployment, and operation, and considering the availability of supporting energy sources. As AI solutions scale, understanding and managing their environmental impact becomes increasingly critical to responsible scale.

Governance of AI solutions would include:

• Appointing a Champion: Health systems could identify a champion for AI – a human leader with passion and leadership to mobilise stakeholders across the sector, that ensures enablers for AI are working as intended, resolving areas of conflict, identifying methods to scale, and sharing leading practices. The Champion could be supported by a series of governance structures designed to oversee the AI in health programme of work.

• Establish accountability and control: To clarify what roles are important to guide the responsible scale of AI solutions and how those roles interact. This would identify clear responsibility for actions and decision making, including which roles are appropriately involved and consulted. It also includes defining controls for analytic uses of data and use of AI solutions – across public health, clinical trials, research, innovation, and health system management. This applies to all stakeholders – across the public, providers, industry, and policymakers among others.

• Establishing incentives to collaborate: Health systems can adopt incentives to encourage collaboration with AI solutions and the use of health data. This would also establish dis-incentives to significantly penalise individuals or organisations who abuse their position to either affect health data privacy breaches or deliberating fragmenting the AI in health ecosystem contrary to agreed objectives.

Emerging leading practices: For post-market surveillance, the OECD has developed an incident reporting framework (OECD, 2025[24]). For further measurement and monitoring, the OECD has developed the Catalogue of Tools and Metrics for Trustworthy AI, including safety, privacy, and fairness (OECD, 2025[25]). In addition, HealthAI is creating a Global Regulatory Network that will include incident reporting (HealthAI, 2026[26]). As work progresses, there will be further opportunities for collaboration on leading practices for governance, measurement, monitoring, and response.

Engagement: Public

Why this is important: The public and the public’s ability to achieve good health outcomes are at the heart of the health system. The public are both the source of most data that are used in AI solutions and the means to measure AI’s effectiveness. As in the OECD Recommendation on Health Data Governance (OECD, 2016[27]), the first component involves the public being involved in the design, implementation, and operation of health data governance solutions (Vanstone et al., 2023[28]). As such, the public should be engaged in all stages of the design, deployment, operation, and improvement of AI in health solutions.

To support meaningful engagement, efforts can also be made to promote digital and AI literacy. Without this, public feedback may be less informed, as such, less valuable, and the public may be less willing to participate. Engagement must also reflect the diversity of the population to ensure inclusive perspectives. Some communities are structurally underserved and may require tailored engagement strategies to ensure their needs and perspectives are reflected in the co-production of AI solutions. Additionally, Indigenous communities hold distinct rights, including in data governance and digital technologies, which must be respected and fulfilled in accordance with relevant legal and ethical frameworks.

Cross-border co‑operation enables countries to draw on shared insights. Because incidents that undermine trust in one jurisdiction may influence perceptions in others, co‑ordinated approaches can help maintain and strengthen public trust.

What this includes: To engage the public, this includes:

• Trust: While trust is critical for the responsible scale of AI in health, there is not yet a consistent definition of what trust requires (Starke et al., 2025[29]). An approach for fostering trust involves engaging with the public to learn what it takes for AI to be trustworthy. This would include transparency in practices and open communication to demonstrate trustworthiness and checking with the public when their requirements for trustworthiness change. Trust can also be strengthened through multistakeholder engagement, ensuring diverse perspectives are included in governance.

• Educate to empower: Improving the awareness, level of digital and AI literacy, and understanding of AI in health will help demystify the new solutions for individuals and allow them to make informed decisions. With the growth of direct-to-consumer access of AI solutions, the public are becoming key actors in their care journey. A focus on education would empower people to be aware of the availability (and trustworthiness) of solutions and how they fit in their overall programme of health and care. This will need to be done along with providers, carefully and respectfully, as this may result in changing the legacy patient-provider dynamic.

• Representativeness: Engagement with the public in all areas above should be designed to be representative – respectful of diversities across cultures, languages, age, and other appropriate socio-demographic factors. In the case of Indigenous Peoples, engagement should respect their data sovereignty, e.g. Canada’s First Nations Peoples of Ownership, Control, Access, and Possession) (FNIGC, 2025[30]); and Australia’s Framework for Governance of Indigenous Data (Commonwealth of Australia, 2024[31])). Working with Indigenous peoples should also respect CARE principles (Collective benefit, Authority to control, Responsibility, and Ethics) (GIDA, 2018[32]). When considering representativeness, there may be value in respecting both individual and collective rights for trustworthy and inclusive AI in health (Alelyani, 2025[33]).

Emerging leading practice: There are several emerging leading practices associated with this area. France has established a Citizen Assembly for Digital Health where diverse members of the public are engaged to discuss issues pertinent to AI and digitalisation of health and making recommendations to the government (Agence du Numérique en Santé, 2025[34]). For education, there are emerging programmes such as Elements of AI, which originated in Finland in 2019, provides free AI education for the public (Elements of AI, 2025[35]).

Engagement: Health providers

Why this is important: Health providers are essential to the functioning of the health system. They serve as a trusted guide to help their patients navigate their health journey. Like the public, it is appropriate for health providers to be involved in all stages of the design, deployment, operation, and improvement of AI in health solutions. As with the public, there is value in being co‑operative across borders and learn from each other and build trustworthiness with the medical community.

What this includes: Engagement of health providers that personally provide care to their patients mirrors that of the public. As such, this includes:

• Trust: Fostering trust involves organisations learning what it takes to be trustworthy, demonstrating that trustworthiness, and checking with health providers when their requirements for trustworthiness change. This is in a continuous, iterative process. Involvement of stakeholders, specifically health providers at every stage of the AI lifecycle is also crucial for building and sustaining trust.

• Educate to empower: Improving the awareness and understanding of AI in health will help to demystify these new solutions for providers to make informed decisions and effectively integrate those solutions into their care delivery workflows. Many AI in health solutions are direct-to-provider. In these cases, health providers should be aware of the reliability and appropriate use parameters of these solutions. Importantly, they should also know how to effectively query and challenge AI outputs when information is inconsistent with their knowledge (Babic et al., 2021[36]). Along with the public, this should be done carefully and respectfully, as this may alter traditional patient-provider dynamics.

• Clinical adoption: Beyond technical assistance and financial investment, adopting AI in healthcare includes clinical workflow redesign, interdisciplinary co‑ordination, and support for leadership structures that can guide responsible implementation. Sustained change management efforts should also address cultural readiness and fostering trust in using AI-enabled tools, supported by transparent communication and clear accountability for outcomes.

Emerging leading practice: There are a number of leading practices in this area, with France mandating AI and digital health training across all health professional programmes starting in the 2025‑2026 academic year (Comité interministériel de l’Intelligence artificielle, 2025[37]), Estonia incorporating AI training in all medical school curricula (TI-Hüpe, 2025[38]), and Germany offering certified AI training through its national KI-Campus platform (KI-Campus, 2022[39]). These examples highlight varying, ongoing efforts to engage and co‑ordinate AI capacity building of healthcare providers.

Engagement: Industry

Why this is important: Industry is a central pillar in the development of AI in health. They are innovators and solution developers to help advance the efficiency and effectiveness of health systems. Industry engagement can aid in understanding existing barriers and opportunities with policy and regulation, co‑investing in evidence generation, and co-creating meaningful solutions that meet the needs of the health system today. It is appropriate for industry partners, which include pharmaceutical and medical technology companies, small, medium and large sized enterprises in health and technology, academia and research institutions, to be involved in all stages of the design, deployment, operation, and improvement of AI in health solutions.

What this includes: Engagement of industry aligns with that of the public and health providers. As such, this includes:

• Trust: Building and maintaining trust requires collaboration to actively cultivate trustworthiness through transparent practices and involve industry through evolving changes. This would focus on predictability of process and minimising unnecessary duplication. Involvement of industry along the AI lifecycle is beneficial for building and sustaining trust.

• Partnership: Developing practices to establish public-private partnerships including the parameters around such partnerships – such as how risks are managed, intellectual property is protected, and how value from the partnership is shared. Could include consideration for embedding members of the public in such partnerships.

Emerging leading practice: There are a number of emerging leading practices in this area, in France, the PECAN programme is offering one‑year of reimbursement for digital medical devices while finalising clinical evidence to obtain certification (G_NIUS, 2025[40]). The United Kingdom has also established the NHS AI lab and associated regulatory sandbox to engage, educate and enable new solutions to enter the health ecosystem (NHS England, 2025[41]). In Australia, the Sparked is actively engaging with industry to enhance interoperability and the foundations for future AI implementations (Sparked, 2025[42]). In Catalonia (Spain), the Observatory of Artificial Intelligence in health provides a transparent repository of all AI tools used in the region, and support for new initiatives to enter the system and scale (AI Observatory in Health, 2025[43]).

Trustworthiness: Responsible Use of AI in health

Why this is important: With AI, humans should feature first and foremost above machines when determining what is ultimately useful for the good of humans. The foundation of this rests in an ethical approach. Within biomedical ethics there are four key principles, as first defined in 1979: autonomy, beneficence, non-maleficence (i.e. ‘do not harm the patient), and justice. (Varkey, 2020[44]). When implementing AI in healthcare, there is an ethical imperative to question and consider the bioethical principles and subsequent ethical impacts of the AI solution.

This ethical imperative extends beyond the clinical setting to overall system decisions about how AI is evaluated, scaled, and governed. There are several ethical principles which can be considered by policymakers, industry, and implementers when embedding ethics into standard procedures. One such guide comes from the OECD AI principles, which provide a global reference point, signalling for AI systems to be inclusive and beneficial to people and the planet; respectful of human rights and democratic values; transparent and explainable; robust, secure, and safe; and accountable (OECD, 2024[4]). These principles reinforce the importance of embedding ethics throughout the entire AI lifecycle, from design, deployment to its use.

As AI is becoming integrated into health systems worldwide, it is critical that countries define ethical principles in alignment with their own societal values, legal frameworks, and healthcare priorities. Countries may develop their own guidelines or adopt internationally recognised tools such as OECD AI principles (OECD, 2024[4]), WHO’s guiding principles for ethical use of AI in health (WHO, 2024[45]), or UNESCO’s recommendation on the Ethics of AI (UNESCO, 2021[46]). These principles guide policymakers and implementers across the AI lifecycle ensuring there are clear ethical guidelines and certainty across the end-to‑end decision making process.

What this includes: Ethical frameworks, principles, and concepts to permeate AI solutions across their lifecycle:

• Ethical development, deployment, and use of AI solutions aligned with applicable legal, regulatory, and societal expectations including boundaries for non-AI development. Provision should be made for human oversight and decision making for scenarios that are not clearly addressed by ethical guidelines. The objective could be to clarify as much as possible to provide certainty for end-to‑end processes while aligning with bioethical principles.

• Bioethical assessment should consider how patient beneficence, patient non-maleficence, patient autonomy, and justice are incorporated into AI solution development, adoption, and scale. For AI solutions intended for the market, the AI solution manufacturer, developer, researcher (and similar) may state how their specific AI solution is aligned with established bioethical principles while quantifying impacts on people and populations with respect to cost-savings, time‑savings, or other meaningful benefits.

Emerging leading practices: International organisations and governments are moving towards the development and use of ethical impact assessment of AI solutions. In New Zealand, Health NZ has established a National Artificial Intelligence and Algorithm Expert Advisory Group (NAIAEAG) as a leading governance mechanism for the ethical and safe use of AI in health. By requiring all AI projects involving Health NZ data or deployment to be registered with the group, and through the use of a structured AI checklist, NAIAEAG ensures that proposed tools meet ethical, technical, clinical and operational standards (Health New Zealand/ Te Whatu Ora, 2025[47]). Taking a different approach, Korea has positioned itself as a global leader in AI governance by integrating ethical oversight into its national legislation. As part of its forthcoming AI Act, expected to be launched in 2026, Korea will establish a national AI committee, chaired by the President and supported by a network of advisory sub-committees. With a five‑year mandate, the Committee is tasked with ensuring that AI governance remains adaptable to technological advancements. The legislation will promote a decentralised ethics model, encouraging public and private sector organisations to establish independent AI ethics committees. (JustAI and Singhal, 2025[48]; Kim & Chang, 2024[49]).

Areas where more convergence is needed: There have been many publications on the ethics of AI, with over 140 documents being published across 19 countries – from governmental organisations, governments, and industry – including aspects of fairness, explainability, and transparency without a clear consensus on requirements (Korukoglu et al., 2025[50]). Ethical guidelines and principles are areas that would benefit from specific action to determine what international consensus is both beneficial and necessary to facilitate the scalability of trustworthy AI in health solutions.

Using the AI in Health Policy Checklist

The above sections have described the policy areas that experts have identified as benefiting from conscious decisions about how they will operate in health systems. These policy areas have been highlighted as they also benefit from consistency within jurisdictions and compatibility across jurisdictions. This approach fosters local autonomy while simplifying the ability to scale solutions. The following table (see Table 2.1) summarises the nine policy areas that make up the checklist for AI in health and suggests key questions to consider for the consistent development of each policy area.

Table 2.1. Policy checklist to support responsible scale and scale of AI in health

Type	Policy Areas	Policy Concept	Consideration – including assuring compatibility across borders were appropriate
Enablers	Better use of health data	Findable	Are there activities that support the findability of health data assets? With national data catalogues?
		Consent for health data	Is there clarity for when consent for health data use is necessary (vs. appropriate alternatives) and how consent is managed?
		Accessible: Primary care	Are comprehensive health data available for the provision of direct patient care?
		Accessible: Analytic use	Are health data available (with protection) for system analytics, public health, research, or innovation with defined controls?
		Interoperable	Are there activities to incentivise and enable the interoperability of data assets?
		Data quality	Are there activities to ensure the integrity of data across organisations? Including timeliness of data for its purpose?
		Linkable	Are there digital identifiers that support linking across data sets?
		Sovereign/secure storage	Are there activities for the secure storage of data assets? Do they include consideration for (indigenous) data sovereignty?
	Enabling the use of AI	Risk management	Are there frameworks for assessment and management of risk associated with AI solutions?
		Technical Assessment	Are there frameworks for the technical classification and assessment of AI solutions? Including update of features or algorithms?
		Algorithmic useability	Are there frameworks to assessment the useability of AI solutions with respect to fairness, explainability, and transparency?
		Model cards	Are there frameworks in use for “model cards” to communicate the result from assessments of AI solutions?
		Procurement	Are there frameworks in use for the procurement of AI in health solutions?
		Consent for AI use in care	Is there clarity for when expressed consent for the use of AI is necessary? And how that consent is managed?
		Approval for AI use in care	Is there an approach that defines boundaries for the clinical use of AI?
		Reimbursement for AI	Is there clarity on how and when use of AI solutions is re‑imbursed and how much?
		Liability for AI use in care	Is there clarity on how liability is managed in care settings?
		Regulatory sandboxes	Are there frameworks for the implementation of regulatory sandbox(es) to accelerate innovation?
	Building capacity and capability	Workforce capacity	Are there actions to develop the evolving health workforce including front-office, back-office and oversight roles?
		Public capacity	Are there actions to build public digital and AI literacy?
		Technical capacity	Are there actions for future technical capacity, including storage and computing power?
Guardrails	Oversight, measurement and monitoring	Post-market effectiveness	Are there actions to measure post-market effectiveness and address issues / incidents as they arise?
		Benefit assessment	Are there actions to measure benefits from the implementation of the AI solution (including consideration for productivity change)?
		Energy usage	Are there actions to measure the energy usage to effectively understand and plan for environmental impact and energy demand?
		Champion / oversight	Is there a designated AI champion for the responsible integration of AI into health systems? Supported by an oversight body?
		Accountability	Is there clarity on end-to‑end accountability and control across the AI life cycle across all identified roles?
		Incentives	Are there incentives in place for collaborative development of AI and use of health data? And penalties for mis-use?
	Agreed objectives	Establish a vision	Is there an endorsed vision for the intended target state?
		Establish an action plan	Is there an endorsed strategy or action plan to achieve the agreed vision?
		Open	Is there consideration for open data and AI solutions as part of the action plan?
		Collaborative	Is there consideration for collaboration as part of the action plan? Does this include cross-border collaboration?
		Responsive / Adaptable	Is there consideration for policies and processes to be responsive / adaptable to risk or opportunity as part of the action plan?
Engagement	Public	Trust	Are there programmes in place for building trustworthiness of AI solutions with the public?
		Educate to empower	Are there programmes in place to educate and empower the public on the use of AI?
		Representative	Are programmes designed to address cultural, linguistic, age, and other diversities? Are considerations for Indigenous Peoples?
	Health Providers	Trust	Are there programmes in place for inclusive design with providers?
		Educate to empower	Are there programmes in place to educate and empower providers on the use of AI?
		Clinical adoption	Are there programmes to optimise adoption of AI tools by providers?
	Industry	Trust	Are there programmes in place for transparent and inclusion of industry in end-to‑end AI life cycle processes?
		Partnership	Are there frameworks to establish public-private partnerships?
Trust	Trustworthy AI	Ethical development	Is there an approach that defines processes to manage ethical considerations for AI solutions? Including boundaries?
		Bioethics assessments	Are there considerations for the alignment with bioethical principles in the development and use of AI?

Note: Bolded items are included in the analysis in Chapter 3.

References

[34] Agence du Numérique en Santé (2025), Assises citoyennes | Agence du Numérique en Santé, https://esante.gouv.fr/assises-citoyennes (accessed on 9 May 2025).

[43] AI Observatory in Health (2025), Observatory of Artificial Intelligence in health - Programa Salut/IA, https://iasalut.cat/en/observatori-ia-en-salut/observatori-dintelligencia-artificial-en-salut/ (accessed on 2 June 2025).

[33] Alelyani, T. (2025), “A validated framework for responsible AI in healthcare autonomous systems”, Scientific Reports 2025 15:1, Vol. 15/1, pp. 44432-, https://doi.org/10.1038/s41598-025-25266-z.

[36] Babic, B. et al. (2021), “Direct-to-consumer medical machine learning and artificial intelligence applications”, Nature Machine Intelligence, Vol. 3/4, pp. 283-287, https://doi.org/10.1038/s42256-021-00331-0.

[17] CHAI (2025), Draft CHAI Applied Model Card - CHAI - Coalition for Health AI, https://chai.org/draft-chai-applied-model-card/ (accessed on 9 May 2025).

[20] Chapman, S. (2025), “Towards identifying good practices in the assessment of digital medical devices: Insights from several OECD countries”, OECD Health Working Papers, No. 177, OECD Publishing, Paris, https://doi.org/10.1787/b485ee1f-en.

[37] Comité interministériel de l’Intelligence artificielle (2025), Faire de la France une puissance de l’IA, https://www.santementale.fr/2025/02/lintelligence-artificielle-au-benefice-du-systeme-de-sante-et-de-ses-acteurs/ (accessed on 29 April 2025).

[31] Commonwealth of Australia (2024), “Framework for Governance of Indigenous Data Acknowledgement of Country Framework for Governance of Indigenous Data”, https://pmc.gov.au/cca. (accessed on 13 May 2025).

[8] Digital Transformations for Health Lab (2025), Data Solidarity Glossary, https://www.dthlab.org/data-solidarity-glossary, https://www.dthlab.org/data-solidarity-glossary (accessed on 15 October 2025).

[35] Elements of AI (2025), Elements of AI free online course, https://www.elementsofai.com/ (accessed on 6 May 2025).

[16] European Commission (2025), CE marking - European Commission, https://single-market-economy.ec.europa.eu/single-market/ce-marking_en (accessed on 9 May 2025).

[30] FNIGC (2025), The First Nations Principles of OCAP® - The First Nations Information Governance Centre, https://fnigc.ca/ocap-training/ (accessed on 13 May 2025).

[40] G_NIUS (2025), Prise en charge anticipée numérique (PECAN) | G_NIUS, https://gnius.esante.gouv.fr/fr/financements/fiches-remboursement/prise-en-charge-anticipee-numerique-pecan (accessed on 6 May 2025).

[32] GIDA (2018), CARE Principles — Global Indigenous Data Alliance, https://www.gida-global.org/care (accessed on 6 March 2026).

[13] Global Digital Health Partnership (2023), “Global Digital Health Model Security Notice (Ver 3)”, https://gdhp.health/wp-content/uploads/2023/06/Proposed-Global-Digital-Health-MSN_CLEAN-Ver3_FINAL-DRAFT_06082023.pdf (accessed on 9 May 2025).

[7] GO FAIR (2021), FAIR Principles, https://www.go-fair.org/fair-principles/ (accessed on 25 August 2025).

[21] GOV.UK (2025), Government Digital and Data Profession Capability Framework - Government Digital and Data Profession Capability Framework, https://ddat-capability-framework.service.gov.uk/ (accessed on 9 May 2025).

[47] Health New Zealand/ Te Whatu Ora (2025), Artificial Intelligence and Algorithm Expert Advisory Group – Health New Zealand | Te Whatu Ora, https://www.tewhatuora.govt.nz/corporate-information/about-us/expert-groups/artificial-intelligence-and-algorithm-expert-advisory-group (accessed on 26 May 2025).

[26] HealthAI (2026), HealthAI — The Global Agency for Responsible AI in Health, https://healthai.agency/ (accessed on 6 March 2026).

[19] IMDRF (2025), International Medical Device Regulators Forum (IMDRF) | International Medical Device Regulators Forum, https://www.imdrf.org/ (accessed on 9 May 2025).

[18] IMDRF (2024), “Medical Device Software: Considerations for Device and Risk Characterization”, https://www.imdrf.org/sites/default/files/2024-02/IMDRFSaMD%20WGN81%20DRAFT%202024%2C%20Medical%20Device%20Software%20Considerations%20for%20Device%20and%20Risk%20Characterization%20-%20final%20draft.pdf (accessed on 26 May 2025).

[48] JustAI and M. Singhal (2025), South Korea AI Regulations, https://justai.in/south-korea-ai-regulations/ (accessed on 26 May 2025).

[39] KI-Campus (2022), Erste zertifizierte Online-Kurse zu KI in der Medizin | AI Campus, https://ki-campus.org/press/zertifizierte-medizin-kurse (accessed on 29 April 2025).

[49] Kim & Chang (2024), Regulations on the Establishment and Operation of the National Artificial Intelligence Committee Enacted - Kim & Chang, https://www.kimchang.com/en/insights/detail.kc?sch_section=4&idx=30261 (accessed on 26 May 2025).

[50] Korukoglu, F. et al. (2025), “Mapping Ethical Guidelines for AI in Healthcare: A Global Perspective”, Studies in health technology and informatics, Vol. 327, https://doi.org/10.3233/SHTI250624.

[14] Lobig, F. et al. (2023), “To pay or not to pay for artificial intelligence applications in radiology”, npj Digital Medicine, Vol. 6/1, pp. 1-9, https://doi.org/10.1038/s41746-023-00861-4.

[41] NHS England (2025), NHS AI Lab, NHS Transformation Directorate, https://transform.england.nhs.uk/ai-lab/ (accessed on 15 April 2025).

[1] Nundy, S., L. Cooper and K. Mate (2022), “The Quintuple Aim for Health Care Improvement”, JAMA, Vol. 327/6, p. 521, https://doi.org/10.1001/jama.2021.25181.

[25] OECD (2025), Catalogue of Tools & Metrics for Trustworthy AI - OECD.AI, https://oecd.ai/en/catalogue/overview (accessed on 6 March 2026).

[11] OECD (2025), Does Healthcare Deliver?: Results from the Patient-Reported Indicator Surveys (PaRIS), OECD Publishing, Paris, https://doi.org/10.1787/c8af05a5-en.

[9] OECD (2025), “Sharing trustworthy AI models with privacy-enhancing technologies”, OECD Artificial Intelligence Papers, No. 38, OECD Publishing, Paris, https://doi.org/10.1787/a266160b-en.

[24] OECD (2025), “Towards a common reporting framework for AI incidents”, OECD Artificial Intelligence Papers No. 34, https://doi.org/10.1787/f326d4ac-en (accessed on 6 March 2026).

[4] OECD (2024), AI principles, https://www.oecd.org/en/topics/ai-principles.html.

[3] OECD (2024), “Governing with Artificial Intelligence: Are governments ready?”, OECD Artificial Intelligence Papers, No. 20, OECD Publishing, Paris, https://doi.org/10.1787/26324bc2-en.

[10] OECD (2023), “Emerging privacy-enhancing technologies: Current regulatory and policy approaches”, OECD Digital Economy Papers, No. 351, OECD Publishing, Paris, https://doi.org/10.1787/bf121be4-en.

[2] OECD (2023), Health at a Glance 2023: OECD Indicators, OECD Publishing, Paris, https://doi.org/10.1787/7a7afb35-en.

[27] OECD (2016), OECD Recommendation on Health Data Governance, https://legalinstruments.oecd.org/en/instruments/OECD-LEGAL-0433.

[22] OECD/LEGAL/0500 (2024), Declaration on Building Better Policies for More Resilient Health Systems, https://legalinstruments.oecd.org/en/instruments/OECD-LEGAL-0500#mainText.

[42] Sparked (2025), Sparked, https://sparked.csiro.au/ (accessed on 6 May 2025).

[29] Starke, G. et al. (2025), “Finding Consensus on Trust in AI in Health Care: Recommendations From a Panel of International Experts”, J Med Internet Res 2025;27:e56306 https://www.jmir.org/2025/1/e56306, Vol. 27/1, p. e56306, https://doi.org/10.2196/56306.

[12] The International Patient Summary (2025), The International Patient Summary – key health data, worldwide, https://international-patient-summary.net/ (accessed on 7 May 2025).

[38] TI-Hüpe (2025), Media – TI-Hüpe, https://www.aileap.ee/media (accessed on 29 April 2025).

[46] UNESCO (2021), Recommendation on the Ethics of AI, https://unesdoc.unesco.org/ark:/48223/pf0000381137_eng (accessed on 6 March 2026).

[15] van Kessel, R. et al. (2023), “Digital Health Reimbursement Strategies of 8 European Countries and Israel: Scoping Review and Policy Mapping”, JMIR mHealth and uHealth, Vol. 11, p. e49003, https://doi.org/10.2196/49003.

[28] Vanstone, M. et al. (2023), “Towards conceptualizing patients as partners in health systems: a systematic review and descriptive synthesis”, Health Research Policy and Systems, Vol. 21/1, https://doi.org/10.1186/s12961-022-00954-8.

[44] Varkey, B. (2020), “Principles of Clinical Ethics and Their Application to Practice”, Medical Principles and Practice, Vol. 30/1, p. 17, https://doi.org/10.1159/000509119.

[5] Weber, C. (2024), Data and trust: the two pillars of value-based healthcare, https://www.weforum.org/agenda/2024/01/value-based-healthcare-data-trust/ (accessed on 9 March 2026).

[45] WHO (2024), “Ethics and governance of artificial intelligence for health Guidance on large multi-modal models”, http://apps.who.int/bookorders. (accessed on 6 March 2026).

[23] WHO (2021), Global strategy on digital health 2020-2025, World Health Organization, Geneva, https://www.who.int/docs/default-source/documents/gs4dhdaa2a9f352b0445bafbc79ca799dce4d.pdf (accessed on 27 September 2024).

[6] World Bank Group (2023), Digital-In-Health: Unlocking the Value for Everyone, https://www.worldbank.org/en/topic/health/publication/digital-in-health-unlocking-the-value-for-everyone (accessed on 9 May 2025).