The World Health Organization's European regional office convened ministers and senior officials from 37 countries in Lisbon on July 15, alongside the Aga Khan University, the Bill and Melinda Gates Foundation, the European Commission, the Wellcome Trust, and the World Bank, for a global conference on AI in health. The statement issued from that meeting names the problem in its title: "Govern AI in health before the gaps become irreversible."
The numbers behind that title are the story. Nearly two-thirds of countries in the WHO European Region are already deploying AI in diagnostics, and half have introduced AI-powered patient chatbots. Only 8% — roughly 1 in 12 — have a health-specific AI strategy in place to govern any of it. The technology is not waiting for the paperwork.
What's live versus what's governed
- Health-specific AI strategy
- 8% of countries
- AI already used in diagnostics
- Nearly two-thirds of countries
- AI patient chatbots live
- About half of countries
- Countries at the Lisbon conference
- 37
Deployment without a rulebook
"AI in diagnostics" and "AI patient chatbots" cover a range of real, already-live tools: imaging-support systems that flag findings for a radiologist to confirm, and triage or intake chatbots that talk to patients before a clinician does. None of that requires a lab breakthrough — it requires procurement, integration, and staff who trust the tool enough to use it. That is precisely why governance lags: deployment is a purchasing decision an individual hospital or ministry can make on its own, while a national AI-health strategy requires the slower work of legislation, liability rules, data protection, and clinician training standards.
Govern AI in health before the gaps become irreversible.
The governance gap, not the model
The open question here isn't whether any given AI tool works — many of these systems are already approved, in use, and presumably delivering value, or hospitals would not have bought them. It's about what happens administratively once a tool is live: who is liable when an AI-supported diagnosis is wrong, how patient data moves through these systems, whether clinicians are trained to override the tool rather than defer to it, and whether a chatbot triaging patients is held to the same standard as the person it replaced. Those are governance questions, and WHO's statement is that most of the region is answering them after the fact rather than before.
What is not established
The 8% and two-thirds figures come from WHO's own member-state survey methodology — self-reported by national governments, not independently audited by WHO or a third party, so they describe what countries say about themselves rather than a verified count. The statement is a call to action from a Lisbon conference, not a binding requirement on any government, and it names no country as having caused patient harm through ungoverned deployment. What it establishes is the shape of the gap, not a case proving the gap has already hurt someone.
- Only 8% of countries in the WHO European Region have a health-specific AI strategy — 1 in 12.
- Nearly two-thirds of the region is already deploying AI in diagnostics; half have AI patient chatbots live.
- 37 countries across all six WHO regions met in Lisbon on July 15 to address the gap.
- WHO's own framing: 'govern AI in health before the gaps become irreversible.'
- Caveat: the deployment figures are self-reported by member states, and the statement is advisory, not binding.