WHO’s New Guidance Puts Ethics Review at the Center of AI Health Research

WHO’s New Guidance Puts Ethics Review at the Center of AI Health Research

September 22, 2026

The World Health Organization has published guidance on ethics review and oversight for artificial intelligence-related health research. The report is directed at researchers, ethics committees, regulators, funders, and policymakers. Its central premise is practical: AI can affect how health studies are designed and interpreted, so existing protections must be applied to the new risks as well as the potential benefits.

AI-enabled research may use patient records, medical images, wearable data, or synthetic datasets. These inputs can be sensitive, and models may reproduce patterns that reflect gaps or bias in the original data. A system’s performance in one hospital or population does not guarantee it will work safely elsewhere.

Ethics review has to follow the whole lifecycle

Researchers should explain what data a system uses, how it was developed, what populations are represented, and how errors will be monitored. Review committees need to understand whether a model is being used to select participants, interpret results, or influence clinical decisions. These roles carry different consequences.

Consent and privacy remain important even when a project uses de-identified data. Re-identification may be possible when datasets are combined, and participants may not expect their information to be used to train commercial systems. Studies should minimize data, define retention periods, and specify whether data or models will be shared.

Fairness and accountability

AI can perform unevenly across age groups, sex, ethnicity, disability, language, or access to care. Researchers should test performance across relevant subgroups and explain limitations. If the sample is too small to support a conclusion for a group, the study should say so instead of implying broad validity.

Accountability also needs a named owner. When an AI-generated output influences a research decision, the protocol should state who reviews it, what happens when the model is uncertain, and how a participant can raise a concern. Human oversight is meaningful only when reviewers have authority and adequate information.

What the guidance means for institutions

Hospitals, universities, and funders can use the report to strengthen review checklists and training. They should ask for model documentation, data provenance, cybersecurity plans, monitoring procedures, and conflict-of-interest disclosures. Post-study monitoring may be necessary when tools continue to be used after a trial ends.

WHO guidance does not automatically change national laws. Its influence will depend on whether regulators and research institutions adapt the recommendations into enforceable procedures. The report is still useful because it makes a clear case that AI health research needs careful ethics review from design through deployment.

Sources

• WHO: Ethics oversight of AI-related health research
• WHO: Ethics and governance of AI for health
• CIOMS International Ethical Guidelines