Blogs

The Organization for Economic Co-Operation and Development (OECD) is an international forum of 38 member countries that sets standards, produces comparative research, and coordinates policy across economic and social domains to support evidence-based policymaking. They produce fundamental resources like the Catalogue of Tools & Metrics for Trustworthy AI and a Regulatory Sandbox Toolkit, which often informs work CHAI does in health AI. Personally, from my roles at the National Health Service (NHS) and beyond, I’ve long-admired the OECD as a highly respected supra-national body and an important vehicle for cross-border alignment.

Since the 2022 OECD meeting, we have worked together to define a shared policy framing for health AI. Naturally, every member country has its own nuances and contexts – e.g. some with Universal Health Coverage, others with employer-based insurance models – and each at a different stage of digital maturity and AI adoption. All are trying to ensure AI is trustworthy, scalable, and implemented effectively.

The OECD's role is to provide a framework that is flexible enough to serve member countries across the spectrum of healthcare models and digital maturity, while being consistent enough to be meaningful. Together, we asked: what are the core principles and standards that can hold across borders, and how do they get translated into country-specific action?

The Event: OECD’s AI in Health Conference

Last month, the Spanish Ministry of Health in collaboration with OECD, convened an international AI in Health Conference in Madrid. The event brought together 142 people, 35 countries and 33 international organizations to agree on an action plan for the responsible adoption of AI in healthcare. The conference was structured around the draft action plan and its three pillars: Trust, Enablement, and Preventing Harm. CHAI led the Enablement track, with our CEO Brian Anderson keynoting and our Director of Operations and General Counsel Brenton Hill facilitating.

The key themes of the event were:

1. AI development is inevitable and arguably urgently needed; governance must keep pace: AI in health is rapidly advancing. 230 million people use AI tools weekly for health-related queries, and AI applications span clinical decision support, diagnostics, rare disease identification, administrative and operational functions, and medical scribes. Meanwhile, there is a shortfall of 11 million health workers globally, concentrated in low- and middle-income countries. Regulatory frameworks, particularly for adaptive and agentic AI, have not kept up. Stage-gated pathways from pilot to practice, with built-in post-deployment monitoring, are urgently needed.

2. AI adoption will move at the speed of trust: Public support for health AI is conditional on human accountability; clinicians need credible, rapid evaluation methods beyond RCTs; and cross-border institutional trust remains fragmented. All three dimensions require deliberate investment.

3. Data governance underpins everything: Health systems hold 30% of the world's data, but legislative gaps, cybersecurity vulnerabilities, and data sovereignty concerns limit safe use. Clear rules enabling trusted secondary use are prerequisites for responsible, trustworthy AI.

4. Equity will not happen by default: Training data skews toward high-income, English-language populations; capital flows away from population health and equity applications; and foundational infrastructure gaps persist in lower-resource settings. Deliberate design choices are required at every stage.

5. International coordination must move from frameworks to action: The ambition is an agreed set of global norms analogous to the Helsinki Protocol, supported by shared indicators, real-world evidence networks, and meaningful inclusion of Global South perspectives — none of which currently exist at adequate scale.

The Action Plan

The Madrid AI Action Plan, which will be announced later this year, will offer an expert-led, internationally-agreed direction for the responsible adoption of AI in health. It will act as a call to action, a trust-building framework, and a set of protective guardrails for the industry.

Following the conference, the OECD is developing a proceedings report to move from principles to prioritization: identifying which actions are most urgent, where existing national and institutional frameworks can be strengthened and built upon, and where the challenges are sufficiently cross-border in nature that transnational cooperation is the only viable path to an effective response.

What this means for CHAI

Just as theHelsinki Declaration was a foundational framework in the 1960s from the World Medical Association that set ethical standards for human medical research – centered on informed consent, risk-benefit proportionality, and participant welfare – the Madrid AI Action Plan intends to be coordinated, actionable commitments that countries “sign on” to and “franchise” their own strategies from.

Think of the Madrid AI Action Plan as the USB-C standard for health AI. It took years of collaboration between competing companies and countries to agree that regardless of who made your phone, where it was manufactured, or what it runs on, it should plug into the same port. That standardization unlocked innovation and made life simpler for users. If we can agree on the core principles and structures that define trustworthy health AI, you can create the conditions for interoperability at scale. Patients shouldn't have to think about the governance infrastructure behind their care. They should simply experience AI that works: safely, consistently, and in their interest, wherever they are.

CHAI is already developing tooling, frameworks, and guidance that maps to the Action Plan, particularly across transparency and explainability, capability-building, trusted data networks, evaluation and assurance, pre- and post-deployment safeguards, and continuous improvement. Working to common definitions and shared high-level principles is how we achieve scale.