The Pulse
WHO urges stronger oversight for AI health research
The World Health Organization has issued a new report urging stronger ethics review for health research involving artificial intelligence. Its recommendations target researchers, ethics committees, regulators, funders and policymakers, with
AI.info Team ·
“At the same time, innovation must be guided by strong ethical safeguards that protect human dignity, rights and equity.”
Dr Meg Doherty, Director, Department of Science for Health, World Health Organization
The World Health Organization is calling for stronger ethics oversight of health research that uses artificial intelligence, warning that existing review systems may not be equipped for the risks created by increasingly complex AI studies.
WHO published its report, Artificial intelligence-related health research: ethics review and oversight, on September 21, 2026. The 69-page document sets out recommendations for researchers, research ethics committees, regulators, funders, scientific journals and policymakers.
WHO separates AI health research into three categories
The report divides AI-related health research into three broad groups. The first covers health research that uses AI to analyze data. The second concerns research conducted with AI tools and technologies. The third examines research on AI tools and technologies themselves.
That distinction matters because the ethical questions differ across each group. A study using an AI model to analyze medical records may raise concerns about privacy, data quality and bias, while a study evaluating an AI diagnostic system may also require scrutiny of performance, accountability and possible harm to patients. Research that develops or tests AI systems can create risks that extend beyond the immediate study and into later clinical or public-health use.
WHO says current oversight mechanisms do not always address concerns involving transparency, fairness, accountability, privacy and the consequences of deploying AI-enabled tools quickly. The report frames ethics review as one part of a broader system that also includes publication standards, data governance, regulation and implementation monitoring.
Ethics committees face new demands
Research ethics committees will continue to protect participants, according to WHO, but many may need additional expertise, training and resources to assess AI-related studies. Committees may need to understand how models are trained, how data are selected, how systems perform across different populations and how researchers will monitor changes after a study begins.
WHO also places responsibility on researchers to identify and reduce ethical risks early rather than treating ethics review as a final administrative step. The report calls for transparent communication, attention to wider social consequences and careful consideration of how AI systems could affect people who are not directly enrolled in a study.
“Effective governance of AI in health research requires collaboration across disciplines, sectors and institutions,” said Dr Alain Labrique, Director of the Department of Data, Digital Health Analytics and Artificial Intelligence at WHO. He said oversight must be “agile, evidence-based and capable of ensuring that innovation remains safe, trustworthy and beneficial for all.”
Oversight cannot stop at the committee door
The report argues that project-by-project ethics review cannot resolve every systemic problem created by AI research. Funders can set conditions for responsible research, journals can demand clearer disclosure of AI methods and data governance bodies can impose safeguards on access to sensitive information.
Regulatory agencies and professional societies also have roles to play, particularly when research findings feed into products or services used in health systems. WHO's approach treats accountability as distributed across the research lifecycle rather than assigned only to the investigators who submit a protocol.
The report also considers the use of AI in ethics review itself. WHO presents potential benefits alongside risks, including the possibility that automated systems could reproduce the same bias, opacity or weak accountability that oversight bodies are supposed to detect. The publication describes its recommendations as non-exhaustive considerations rather than a final standard.
Low-income countries face unequal exposure to AI risks
WHO gives particular attention to low- and middle-income countries, where AI could help address persistent health challenges but where research capacity and decision-making power remain unevenly distributed. Much of the current research and technology development is concentrated in higher-income settings, the report says.
That imbalance can affect whose data are used, who defines research priorities and who receives the benefits of resulting systems. WHO calls for stronger local leadership, capacity and participation so that countries are not treated only as sources of data or testing environments for tools designed elsewhere.
The report identifies equity concerns that include power imbalances, unfair distribution of benefits, data colonialism and ethics dumping. Those issues cannot be resolved solely by adding a technical specialist to an ethics committee; they require institutions to examine who controls the research, who bears its risks and who can challenge its results.
A starting point, not a finished rulebook
WHO presents the publication as a starting point for developing future standards, guidance and governance approaches for AI-related health research. Its recommendations do not create a single global approval process, and the report does not claim that one review model will fit every type of study or health system.
The immediate message is narrower and more practical: AI research should receive oversight that matches its methods, potential effects and reach. For WHO, that means giving ethics committees the expertise and resources to review complex studies while requiring funders, publishers, regulators and data institutions to carry responsibility beyond the initial research application.