Artificial intelligence (AI) in health is rapidly advancing: an estimated 230 million people use AI tools weekly for health-related queries and 35% of the public report using AI before they meet a health provider. AI applications are increasing in scope, spanning clinical decision support, diagnostics, public health system management, as well as administrative and operational functions such as medical scribes, and other areas.
Given the rapid deployment of AI and its demonstrated advantages for specific tasks, the question is no longer “should we adopt AI in health?”, but “how do we adopt AI effectively and safely?”. This marks an inflection point where the mindset around AI is moving from experimentation to scale.
However, health systems are struggling to scale AI in health efficiently and effectively. More than 90% of executives report no measurable impact on productivity. There are numerous emerging principles, guidance, and frameworks for how to achieve value; however, these are often fragmented and built independently. This creates collective noise for policymakers with inconsistent language, priorities, and framing.
Inspired by the drive to accelerate the effective scaling of AI in health systems for better human and economic benefits, the OECD and the Spanish Ministry of Health co‑organised a conference on the “Responsible Scale of AI in health”. The conference was held on 28‑29 May 2026 in Madrid, Spain, and brought together 142 experts, from 26 OECD Members, 8 accession candidate and Key Partner countries, and 35 international partners.
Through the conference, several themes became apparent; (1) scale happens at the speed of trust, (2) foundations of digital and data infrastructure, policy, oversight, technology lifecycle management would benefit from greater coherence, and (3) economic and financial costs and benefits for the use of AI in health need to be better understood.
Experts were asked to identify actions where collaborative guidance would be most beneficial for health systems to advance the effective scale of AI in health for human and economic benefit. They were further asked how to create a “signal above the noise” – that is, incentives and enablers – for policymakers and industry. The intended outcomes from the actions would be to reduce unnecessary redundancy by fostering strong foundations including assessment of the impact of AI initiatives in real-world settings, providing economic and financial models that support scale of valuable initiatives, and supporting innovators to design, develop, and deploy them with more certainty. Further actions would help to foster trust and use of AI among health providers and the public. Experts were asked for existing initiatives and actions to build on what is working, so that it can be amplified, or identify gaps that could be addressed collectively.
These themes formed the basis for the Expert-Led Madrid action plan (EL-MAP). In this proceedings report, countries and experts identified existing initiatives that will support the implementation of EL-MAP.
Experts identified ten actions for EL-MAP. The actions are aligned with the themes from the Conference.