Proceeding report from Madrid conference on the responsible scale of AI in healthcare

3. Madrid Conference proceedings

Madrid Conference for the responsible scale of AI in health

In partnership with the Ministry of Health of Spain, the OECD convened an international conference on scaling AI in health in Madrid on 28‑29 May 2026. The conference was attended by 142 experts, from 26 OECD Members, 8 accession and partner countries, and 35 international partners (see 4Annex A). Participants were nominated through various channels with the aim of ensuring multi-disciplinary and geographic representation. Collectively, experts brought expertise from academia, industry, public sector, civil society, and patients’ and clinicians’ representatives.

Programme and working sessions

Coinciding with broader international efforts to preserve a human-centred approach to AI’s development, the conference was organised around three sessions that explored fundamental areas to earn trust, enable what works and prevent harm. To incentivise knowledge sharing and group discussions, the second and third sessions were organised as breakout groups, during which the experts discussed challenges and leading practices. This resulted in expert advice on the urgent actions required to effectively scale AI in health.

Opening session: Setting the stage

The conference opened with remarks from Spain’s Secretary of State for Health (Javier Padilla Bernáldez) and the OECD’s Deputy Secretary General (František Ružička) as well as pre‑recorded addresses from the Health Ministers from South Korea (Jeong Eun-kyeong) and Australia (Mark Butler). Across the remarks, the objective was set, summarising the challenges as described in Section 1 (see Beyond technology: Why AI struggles to scale in health.

Collectively, the speakers outlined the desired outcomes of the Conference:

• Build a shared understanding of gaps in scaling AI in health;

• Define a collective vision of AI-enabled health systems;

• Surface what experts are doing and where transferable practices exist;

• Agree on foundations for a concrete action agenda.

Collectively, the group was tasked with formulating an expert-led Action Plan. This would provide a clear signal for action to reduce fragmented guidance on the effective scale of AI in health.

First working session: Earning Trust

The first session focussed on Earning Trust in AI systems sharing perspectives from the public (including youth), patients, providers, health systems and industry on the future of health systems. The session featured a keynote address from Jennifer Dixon (CEO, The Health Foundation) and the release of a paper series on AI governance for young people from the Digital Transformation for Health Lab (Digital Transformations for Health Lab, 2026[46]).

There were several key takeaways from the first session by participants. First, the effective scale of AI will happen at the speed of trust (Binesmael et al., 2026[47]). That is, there is a need for increased involvement of patients and providers in the design and deployment of solutions, including the need for formal “change management” programmes. The French Citizen Assembly for the Digitalisation of Health (Assises Citoyennes du Numérique en Santé) and the Canada Declaration of Personal Health Data Rights were cited as examples on which to support trust-building (Ministère de la santé, 2025[48]; Patients' Annual Summit, 2026[49]). It was also noted that foundational medical ethical and human rights principles – principles that help to foster trust – already exist. There is no need to reinvent ethical principles – rather the challenge is applying them in practice.

Second is that the public are already direct users of AI solutions. 35% of the public across 16 countries report using AI for health purposes before consulting a doctor (Edelman Trust Institute, 2026[9]). Action is necessary to support public use of these systems by improving public awareness, access, and understanding of these platforms. The outcome is for the patient to be able to safely use AI solutions as part of their overall health journey. Programs like Finland’s Elements of AI (a free on-line course to learn about AI) was given as an example of how to address this area (Elements of AI, 2024[50]).

A surprising finding from the discussion is that youth populations are less keen on AI-based solutions, given their higher exposure to news and their concerns about job displacement (Thornton et al., 2024[51]). Despite that, few strategies acknowledge youth populations as requiring special consideration. The message from the experts was that if we want to secure the future, there is need to focus more on the implications of AI for youth populations.

In addition, leveraging AI for public health purposes was identified as a blind spot throughout the discussions. While AI has the potential to support disease surveillance, health promotion, and population-level interventions, most applications remain focussed on healthcare delivery, such as diagnostics and clinical decision support. It was also highlighted that algorithmic amplification, where AI solutions amplify the most observed – not necessarily the most accurate – information, through the maximisation of young people’s engagement, poses a real public health issue (Mori, 2025[52]). Platforms can inadvertently promote misleading health content to the detriment of young people’s health (Digital Transformations for Health Lab, 2026[46]).

There was also a call to help design the workforce of the future to support an orderly transition. It was acknowledged that, in the future, the traditional roles of doctors and nurses will transition and be better integrated with AI. New specialist and hybrid positions will also be created, and that there will be a need to invest more in supporting functions (e.g. developers, data architects) so that the front-line uses of AI can deliver optimal benefits. Remuneration models would also need to adapt, requiring collaboration across stakeholders to develop approaches and advance this work.

Finally, the necessary role of the private sector was acknowledged as a key partner in implementation of AI in health. Industry can contribute expertise, investment and implementation capacity, while governments, providers and the public help define health system needs, safeguards and public assets, including protected health data. These collectively provide the enabling conditions for innovation.

Second working session: Enabling What Works

The second session focussed on Enabling What Works, reflecting that there is significant expertise in health in integrating novel technologies. The session featured a keynote address from Brian Anderson (CEO, Coalition for Health AI) and four parallel breakouts to discuss actions to (a) establish trusted data and digital collaboratives; (b) align needs, opportunities, and evaluation; (c) accelerate integration and scale; and (d) enable competitiveness and adaptability. There were several critical insights from the discussions.

First was that the future of AI will be built on a strong foundation of interoperable and accessible data. Experts reflected that health is at a point where the harm resulting from the non-use of health data may outweigh the potential harm from its use. Further, divergence in approaches to health data governance and digital infrastructure negatively impact the ability to collaborate, innovate, and scale. Experts called for clarity on valid secondary uses of health data and tools to simplify how and when data are made available for those purposes while maintaining strong trust, consent, privacy, and cyber security protections. This echoes OECD work highlighting the need for greater convergence of (health data) governance frameworks across jurisdictions and harmonisation of national procedures for the secondary use of health data for public-interest purposes (OECD, 2025[53]), as well as the OECD Recommendation on Health Data Governance, which calls for processes that support privacy-protective data access and use (OECD, 2016[54]). Participants also noted Canada’s emerging framework on data-related harms (Government of Canada, 2026[55]), and the European Health Data Space Regulation as examples of efforts to clarify and enable secondary uses of health data (European Commission, 2025[56]).

Second is to assure that the right incentives are in place to support a shift from technology-centric innovation to needs-based innovation with an emphasis on enabling data re‑use. It was acknowledged that while pilots may be useful, those without a clear path of how they would scale efficiently and effectively are wasteful activities. Several countries (France, Lithuania, Norway, Spain) noted examples of how they are using incentives differently to achieve this aim. This should be supported by an economic analysis of the potential economic and human benefit from AI in health – and what it would take to get there.

Among other insights, the group also noted the use of regulatory sandboxes to help accelerate innovation. These tools allow for innovation in a controlled space to understand the interaction between novel technologies and legislation, to assure the safety and effectiveness of solutions. Such sandboxes are already in place in the United Kingdom, the United States, and several other countries. This was noted as a capability that could benefit from scale.

Third working session: Preventing Harm

The third session focussed on preventing harm, reflecting the human aspect of health where innovation can both cause and prevent better health outcomes. The session explored actions that could optimise human and economic outcomes while protecting patients, providers, and systems from harm. The session featured a keynote address from Ricardo Baptista-Leite (CEO, HealthAI – the Global Agency for Responsible AI in Health) and four parallel breakouts to discuss actions to (a) embed privacy and security by design; (b) preserve human oversight; (c) support pre‑ and post-deployment safeguards; and (d) facilitate continuous learning. There were several critical insights from the discussions.

First, there is a need for leadership among health systems to listen to and incorporate various perspectives across industry, developers, policymakers, and the public. While many countries are developing approaches to AI governance in health, there is considerable variation in the institutional arrangements, oversight mechanisms and processes used to guide the development and deployment of AI, as well as in how their effectiveness is assessed at local, regional and national levels. There is an opportunity to collaborate and share insights in how the governance of AI is operated and measured.

Further, progress was noted on pre‑market authorisation and post-deployment monitoring of AI solutions, learning from approaches used for medical devices or pharmaceuticals to detect and address safety issues quickly and effectively. The group noted significant progress being made by HealthAI with their initial set of Pioneer countries in their Global Regulatory Network (HealthAI - the Global Agency for Responsible AI in Health, 2026[57]), where participating countries commit to strengthening governance capacity for AI in health solutions and co-developing a Global Public Directory of registered AI solutions in health as well as an Early Warning System for adverse events associated with deployment.

The panel also noted the need to strengthen cyber-security protections. Cyber-attacks are increasingly becoming more sophisticated and can be characterised as a technological “arms race” where AI is both being used to enhance attacks and to strengthen defensive capabilities to protect against attacks. The potential emergence and wider adoption of quantum computing could create both opportunity and risks for health systems. Preparing human networks to share insights and mutually strengthen cyber-resilience was noted as an urgent area of action.

Closing panel and remarks

The penultimate session featured a panel of high-level speakers from a spectrum of geographical regions (Europe, Latin America and Africa). The panel reflected on:

• How this effort would help accelerate the effective scale of AI through better co‑ordination across regional and international networks;

• How this timely Expert-Led Madrid Action Plan could act as an enabler in translating general aspirations into a more technical and governance‑oriented conversation to effectively scale AI in health;

• How the Expert-Led Madrid Action Plan would help provide a clear signal for collaboration while allowing regions and countries to tune it to local needs; and

• How a collaborative approach would be helpful for scalability and trans-national collaboration.

In the final session, the OECD’s Chair of the Health Committee (Chris Mullin) and Spain’s Minister of Health (Mónica García Gómez) thanked participants and organisers while expressing optimism about next steps in collective efforts.

Key themes and identified actions by the experts

In summary, across the sessions and discussions, experts were able to link the key challenges raised by participants with the desired outcomes and the actions required to achieve those outcomes. Themes can be summarised in three parts:

• Establish and sustain trust

• Strengthen foundations

• Develop economic and financial models

Establish and sustain trust

Experts noted that scale only happens at the speed of trust. All health system actors need to be considered as well as their interactions, particularly as health systems undergo major transformation that needs to be anchored in a human-first approach. There were three main areas that would benefit from a collective approach.

First is co-design of use cases that would reflect not only where AI could be integrated into health systems, but also which needs it should address and what desirable solutions should look like. This would require health systems, practitioners, patients, policymakers, payers and developers to collectively define the problems to be solved, the outcomes to be improved, and the care pathways or public health functions where AI could add greatest value. This approach would learn from the mistakes of the initial waves of the digitalisation of health that delivered digitised solutions such as scanning paper records or moving services to the cloud, rather than digitalising health by re‑designing workflows and care models with health practitioners and patients. Those mistakes fragmented health data, added administrative burden to practitioners, and failed to realise the promise of digitalisation. Co-designed use cases could provide a clearer focus for collaborative action to scale up solutions with the greatest human and economic value.

Second, experts noted a need for capacity building and upskilling among users (including the public, patients and healthcare professionals) to understand how their role would contribute to the effective scaling of AI in health. While this is often described as co-building AI literacy, it also requires strengthening data and digital literacy to optimise effectiveness. Experts noted that this would go beyond patient-facing uses of AI (including building capacity among the public) to roles involved with the procurement, management, and oversight of AI systems across primary and secondary uses of AI. Experts noted that building capacity is not enough – there is an opportunity to work with the AI development community to simplify the use of AI for health in ways that foster clarity, use, safety and trust.

Third, experts reflected on the opportunity for better collaboration across sectors and stakeholders to bring forth the human and economic value of AI in health. The public sector is often the steward of health system data and digital infrastructure necessary for effective AI in health solutions. The private sector often has the resources and risk structures to develop those AI in health solutions. Civil society (including providers and the public) can help provide guidance for organisations or collaborations to foster trust in their design and execution.

Strengthen foundations

“For AI to be useful, it must be supported by an enabling environment” (United Nations, 2026[58]). In shifting from experimentation to scale, innovators integrate their solutions into the health ecosystem. This is similar to the approach in urban planning, where governments provide common foundations – such as electricity and water – that allow individual builders to leverage those foundations. This lowers cost for builders accelerates their ability to scale. There are also benefits for the community from upgrades to the foundational infrastructure (such as wastewater improvements). The effective scale of AI would benefit from a similar approach. There is opportunity to identify and strengthen the modern common foundations of health so innovators can take advantage of those when moving from experimentation to scale.

There are several common foundations for AI in health. A co‑ordinated policy environment provides clarity in several areas – access to training data, methods for technology evaluation, and requirements for consent and data governance among others. Leveraging common (data and digital) infrastructure helps solutions to be interoperable, leveraging technology (e.g. cloud) services, and supports data to be used for care, planning, public health, or innovation with appropriate protections. Following a common approach for AI life cycle management – learning from the software development life cycle – will support clarity for developers from design through development, evaluation, deployment, post-market monitoring, and subsequent improvement.

Another foundational aspect for AI in health is leadership – the role that helps to assure clarity around accountability, addressing barriers and risks to adoption, providing incentives to align with foundations among other areas, and encouraging collaboration toward common objectives. Leadership can help transform the adoption of AI in health from a series of isolated innovations competing for limited resources to a co‑ordinated set of activities that are collectively designed to achieve the desired human and economic outcomes.

Finally, AI in health will benefit from supporting adaptability for emerging technologies and opportunities. Experts reflected that change is the new normal. Solutions should be designed to adapt to new capabilities (e.g. precision medicine) and threats (e.g. cyber-attacks). This could impact infrastructure design, policy structures, and leadership incentives.

Develop economic and financial models

The short-, medium-, and long-term human and economic benefits from AI should be understood, along with the incentives that would help to effectively and efficiently deliver sustained value.

Experts reflected that developing an economic model for AI in health will help support business cases for investment. This would benefit from improved measurement of the impact of AI in health solutions including through Health Technology Assessments and cost-effectiveness studies, considering the time frame for deployment, when benefits would accrue, and to whom.

This economic model would be accompanied by a model for financing AI in health. This could include incentives for the adoption and sustained use of AI in health, impacts on remuneration linked to the use of AI, and methods to cost- and value‑share in the building of solutions that consider value, risk, and expenditure. It could also consider how the underlying business architecture of health systems may need to adapt so that reimbursement, procurement, workflow design and accountability arrangements incentivise the adoption of validated AI tools that generate measurable human and economic value.

An economic model would incorporate the economic benefits of re‑using common foundations (as described above). Experts reflected that investing in common foundations (data and digital infrastructure, policy, lifecycle management, leadership, and adaptable capabilities) would create economies of scale by spreading substantial upfront investment costs across multiple AI applications. However, the economic model should distinguish between capital investment required to establish common foundations and the recurring costs required to operate, maintain, govern, and adapt them over time.

Shared infrastructure models, such as digital public infrastructure and telecom tower-sharing arrangements, demonstrate how upfront investments can be pooled while ongoing operating costs can be sustainably distributed among users. Applying a similar approach to AI in health could reduce duplication, lower the marginal cost of deploying additional AI use cases, and create more sustainable pathways for scaling AI across health systems.

This could be particularly valuable for low- and middle‑income countries, where high upfront investment costs and limited operational resources often constrain adoption, and for the development organisations that support them. Stronger common foundations would also facilitate the re‑use of data in the public interest, creating further opportunities to increase both the economic and societal returns on AI investment.

Toward an expert-led action plan

Over the Conference, in addition to identifying opportunities for collaboration, experts also provided examples of existing initiatives that are already addressing these themes. This demonstrates that there is real opportunity to accelerate action to build trust, strengthen foundations, and provide an economic basis for AI in health by working together. Many of the identified themes are already being addressed by emerging policies, governance frameworks, implementation tools, and collaborative initiatives worldwide. These existing activities provide valuable examples that can be adapted, expanded, and co‑ordinated across experts and organisations to accelerate the effective scaling of AI in health. This would give guidance of emerging leading practices, while not being prescriptive.

These reflections informed the drafting of the expert-led Madrid Action Plan (EL-MAP) in the next section.