The World Health Organization's main statement on AI ethics is its guidance "Ethics and governance of artificial intelligence for health," published on June 28, 2021, which sets out six consensus principles for AI used in health care and public health. A January 18, 2024 follow-up applied those principles to large multi-modal models, the systems behind many chatbots. As of October 2026, WHO has not replaced the six principles, but it has added newer papers on research ethics and on AI in health policy. All of these are guidance for governments, developers, and health systems, so they shape policy debates without being binding law.
Key takeaways
- The 2021 guidance was prepared with a group of twenty experts and describes its six principles as the first consensus principles in this field.
- The principles are: protect autonomy; promote well-being, safety, and the public interest; ensure transparency, explainability, and intelligibility; build responsibility and accountability; ensure inclusiveness and equity; and promote AI that is responsive and sustainable.
- The 2024 guidance on large multi-modal models contains more than 40 recommendations and splits responsibility across developers, providers, and deployers; on liability it lists a presumption of causality, strict liability, or no-fault compensation funds.
- In 2026 WHO added a September report on ethics review of AI-related health research and a June discussion paper on AI in evidence-informed policy; neither replaces the 2021 principles.
- WHO guidance does not create legal duties in Canada or the United States, where drug, device, and privacy regulators set the binding rules.
The 2021 guidance and who wrote it
WHO's 2021 guidance came from two WHO departments, Digital Health and Innovation and Research for Health, and an external expert group co-chaired by Effy Vayena and Partha Majumder. The foreword by WHO's then Chief Scientist, Soumya Swaminathan, describes a two-year development process and notes that AI could help patients take control of their care but could also move decisions that belong to providers and patients over to machines. In the document's wording, autonomy means that humans should remain in full control of health-care systems and medical decisions.
The document also covers ethical challenges such as the digital divide, bias, and commercialization, along with impact assessment and liability regimes. The six principles are its core.
The six principles and what each asks for
Each principle in the 2021 text comes with specific expectations. The table pairs them with a short reading of what a hospital, regulator, or developer would be expected to show. The right-hand column is a summary of the guidance, not an official checklist.
| Principle | What the guidance asks for |
|---|---|
| Protect autonomy | Meaningful human oversight, the ability for a clinician to override the system, and informed consent that covers the use of machine learning |
| Promote well-being, safety, and the public interest | Meeting safety, accuracy, and efficacy requirements before deployment, then continuous measurement of performance and harm |
| Ensure transparency, explainability, and intelligibility | Published or documented information on assumptions, limits, data, and the model, enough for audits and public debate |
| Responsibility and accountability | Clear task specifications, points of human supervision, and redress for people harmed by algorithmic decisions |
| Ensure inclusiveness and equity | Diverse design teams, testing across groups, and attention to bias, language, and the digital divide between countries |
| Promote responsive and sustainable AI | Ongoing review of whether a tool works as promised, a process to stop using it if not, and attention to energy use and workforce effects |
A concrete example in the text comes from the equity section: a skin-lesion diagnostic trained on data from one skin color may give inaccurate results for patients with another, which raises their health risk. The responsiveness section adds that new technology in under-resourced health systems is often not repaired or updated, wasting resources that could have gone to proven interventions.
The 2024 follow-up on large multi-modal models
On January 18, 2024, WHO issued guidance on large multi-modal models (LMMs), which accept inputs such as text, images, or video and produce varied outputs. It names five uses in health: diagnosis and clinical care, patient-guided use, clerical and administrative tasks, medical and nursing education, and scientific research and drug development. The listed risks include false, biased, or incomplete statements, poor-quality training data, unequal access, automation bias, and cybersecurity weaknesses that could expose patient information.
Its main structural idea is the AI value chain, which separates developers who build a general-purpose model, providers who adapt it into a product, and deployers such as ministries of health, hospitals, and individual clinicians who use it. The document says governments define obligations at each stage, while developers hold responsibility during design. Among its recommendations for governments are assigning a regulatory agency to assess and approve LMMs for health use, as resources allow, and requiring independent post-release audits and impact assessments for large-scale deployments, with results broken down by user characteristics such as age, race, or disability.
Liability options the 2024 text lists
The 2024 document says errors, misuse, and harm to individuals are inevitable as LMMs spread. It suggests that governments could introduce a presumption of causality, which would ease the burden of proof for a harmed person, or a strict liability standard for harm from deployment. It also notes that strict liability might discourage the use of more sophisticated systems, and mentions no-fault, no-liability compensation funds as a further option. These are options for lawmakers to weigh, not rules in force.
What WHO added in 2026
The 2021 principles still stand as WHO's core statement, and WHO's publication page for them lists no newer edition. Two 2026 documents build on the same ground. On September 21, 2026, WHO launched "Artificial intelligence-related health research: ethics review and oversight." It sorts AI-related research into three groups: studies that use AI-analyzed data, studies that use AI tools, and studies about AI tools themselves. It recommends that ethics oversight cover the whole research lifecycle and that research ethics committees get more expertise, training, and resources. WHO calls the report a starting point for future standards and guidance.
Earlier, on June 2, 2026, WHO published a discussion paper, "Artificial intelligence and evidence-informed policy: emerging challenges and opportunities." It suggests algorithmic impact assessments before deployment and human decision gateways during use, and it says AI should support human decision-makers rather than replace them. The paper also recommends adapting existing instruments, including WHO's AI ethics guidance, instead of writing new frameworks. These are discussion material for Member States, not requirements.
What the guidance does and does not do
WHO guidance carries weight because it reflects the consensus of an international expert group. It does not replace national law. In Canada, medical software licensing and privacy duties come from federal and provincial rules, and in the United States from agencies and statutes, so a hospital cannot meet its legal duties by pointing to a WHO principle. Each document describes risks and expectations as of its publication date, and none tests a particular product. This article gives general information, not legal or medical advice, and a specific tool or decision calls for a qualified professional and the original documents.
FAQ
Is the WHO guidance a law or a treaty?
No. It is guidance addressed to governments, developers, and health systems. Binding requirements come from national and regional law.
How many principles are in the WHO guidance on AI for health?
Six. They appear in the June 28, 2021 guidance and are repeated in the January 18, 2024 document on large multi-modal models.
Does the WHO approve or rank specific AI tools?
The two documents examined here do not rate or approve products. They set out principles, risks, and governance recommendations.
What did the 2024 guidance add to the 2021 principles?
It applied the principles to LMMs, added a developer, provider, and deployer view of responsibility, and listed options for liability and international governance. It also lists more than 40 recommendations.
Has WHO issued anything newer on AI ethics in health?
Yes, in 2026: a June discussion paper on AI and evidence-informed policy and a September report on ethics review of AI-related health research. Neither revises the six principles.
Related reading
- Controversial debate topics in AI
- Ethical dilemmas AI presents
- Ensuring morality in AI systems
- Why AI hallucinations matter
More in AI Policy and Ethics.
Last verified: October 2026
Sources
- Ethics and governance of artificial intelligence for health: WHO guidance, World Health Organization, June 28, 2021
- Ethics and governance of artificial intelligence for health: Guidance on large multi-modal models, World Health Organization, 2024
- WHO releases AI ethics and governance guidance for large multi-modal models, World Health Organization, January 18, 2024
- New WHO report calls for stronger ethics oversight of AI-related health research, World Health Organization, September 21, 2026
- New WHO discussion paper sets out opportunities and risks of AI in evidence-informed health policy, World Health Organization, June 2, 2026