For the past several years, conversations about artificial intelligence in healthcare have often circled the same well-worn territory: sweeping statements about AI’s transformative potential, careful discussions of ethical guardrails, and an abundance of pilot projects that rarely seem to graduate into sustained, real-world use. This week in Hangzhou, China, that conversation took a deliberate turn toward something more concrete, as the third meeting of the Global Initiative on AI for Health convened with an explicit, stated goal: shifting the global AI-in-healthcare conversation from principles toward actual implementation.

The three-day meeting, running from September 16 to 18, was jointly convened by three United Nations specialized agencies whose combined expertise reflects just how multidimensional the challenge of responsibly deploying AI in healthcare has become: the International Telecommunication Union, which drives global standards and connectivity for AI-enabled technologies; the World Health Organization, which provides authoritative guidance on the safe and ethical use of AI within health systems; and the World Intellectual Property Organization, which serves as the leading international forum on the intersection of AI and intellectual property. Together, these three organizations formed the Global Initiative on AI for Health, known by the shorthand GI-AI4H, back in 2023 — and this week’s gathering marked only the third time the full initiative has convened since its founding.

The meeting brought together an unusually broad cross-section of stakeholders: policymakers, regulators, health practitioners, researchers, technical experts, and representatives from both civil society and industry, all working through an agenda that organizers described as centered on putting responsible AI for health into practice. That emphasis on practical implementation, rather than continued abstract debate about AI governance principles, reflects a broader and increasingly urgent recognition within global health circles that AI tools are already being deployed across health systems worldwide — with or without the kind of coordinated international guidance and standards that bodies like GI-AI4H exist specifically to develop.

A central component of this week’s proceedings involved reporting interim results from the GI-AI4H Benchmarking Challenge, an ambitious open global initiative designed to evaluate AI health models against real clinical data without requiring either party in the evaluation — the model’s developers or the institution holding the clinical data — to fully disclose their respective proprietary assets to each other. The challenge represents a technically sophisticated attempt to solve one of the thorniest practical problems in healthcare AI development: how to rigorously test AI models against genuine clinical data, which is often subject to strict privacy protections and institutional ownership constraints, without compromising either the security of patient data or the intellectual property of the AI developers whose models are being evaluated. Organizers were careful to note that the benchmarking results reported at this week’s meeting do not themselves constitute regulatory approval, certification, or endorsement of any specific AI model — the challenge is designed purely as an evaluation mechanism, distinct from any formal regulatory clearance process that individual countries or health systems might separately require.

Beyond the benchmarking results, this week’s meeting also featured new guidance jointly developed by WHO, the ITU, and WIPO specifically addressing how AI-enabled health innovations can move from initial development through to real-world deployment and eventual scale — guidance that reportedly places particular emphasis on intellectual property considerations throughout that lifecycle, alongside case studies drawn from companies operating in different healthcare and technology markets around the world. The publication reportedly features what organizers describe as an “Innovation Lifecycle IP Matrix,” intended to help map strategic intellectual property decisions from a technology’s earliest conceptual stages through to global-scale deployment.

Organizers also specifically invited participating organizations to submit real-world examples of AI applications already deployed within health services or broader health systems, with particular interest in submissions describing applications contributing to primary healthcare delivery, progress toward universal health coverage, or broader health-system strengthening efforts. Notably, the submission criteria reportedly required more than a simple technical description of each AI application — organizations were asked to detail the scale at which their AI tools have actually been deployed, the concrete results achieved, the supporting evidence base underlying those results, and the governance arrangements put in place to oversee the technology’s use, a set of requirements that reflects a deliberate effort to filter out purely aspirational or pilot-stage projects in favor of genuinely operational, evidence-backed AI health deployments.

The location of this week’s meeting — Hangzhou, China — carries its own layer of significance, given China’s increasingly prominent role in global AI development more broadly and its own substantial, rapidly scaling investments in AI-enabled healthcare technology across its domestic health system. Hosting GI-AI4H’s third global meeting in China offered international stakeholders a chance to engage directly with Chinese health AI deployments and policy approaches, at a moment when questions about how different national regulatory environments will shape the global trajectory of healthcare AI remain very much unresolved.

As the meeting’s three days of sessions concluded, the underlying message from organizers was consistent and, by the standards of past AI-in-healthcare conversations, notably direct: simply demonstrating that AI tools can work in healthcare settings is, at this point, no longer sufficient. The harder, more consequential work now lies in scaling those tools responsibly, governing their use effectively across dramatically different health systems and regulatory environments, and ensuring that the benefits of AI-enabled healthcare reach beyond the well-resourced institutions and countries currently best positioned to develop and deploy them — a goal that this week’s meeting in Hangzhou took concrete, if incremental, steps toward advancing.