Public health institutions should treat the task of distinguishing what is known from what is merely claimed as a core responsibility, according to Professor Kent Buse and Professor Sarah Hawkes of Monash University Malaysia.

In an Opinion article published recently in PLOS Global Public Health, the authors warn that the institutions responsible for turning information into evidence are being defunded and discredited just as generative artificial intelligence (AI) makes plausible-sounding figures and claims instant and virtually free to produce.

The article, titled “Who sifts? Knowledge as a public good in the age of AI,” argues that public health agencies must strengthen their systems for checking, documenting and defending evidence.

A fragile foundation for global health data

Buse and Hawkes point to the US-funded Demographic and Health Surveys programme, which for four decades provided data on population trends, child mortality, nutrition, immunisation and other health measures in more than 90 countries.

Ministries, United Nations agencies and researchers relied on the programme’s findings to guide policy and assess progress. Yet the United States ended its funding in February 2025, despite the programme costing less than 0.1% of the former USAID budget.

AI tools can present a measured statistic, a modelled estimate, a figure from a press release and an unsupported or fabricated claim in the same fluent style. They can also produce inaccurate or invented citations

Surveys were under way or ready to begin in 23 countries when funding stopped. Private philanthropy has since supported a reduced programme, but the authors warn that a major source of global health data now rests on a limited-term funding arrangement.

The vulnerability of existing information systems was illustrated while the authors were preparing their article. When they checked the status of the Demographic and Health Surveys programme, the first answer returned was an AI-generated summary. It was confident, but only partly sourced.

AI increases pressure on evidence systems

Generative AI is also being introduced into the organisations responsible for assessing and communicating health information.

By late 2024, the UN system had catalogued more than 700 AI projects across approximately 50 entities. Almost half involved text or language applications. The World Health Organization launched a generative health chatbot in 2024, although an early assessment identified operational problems that could damage confidence in the service.

The authors say AI tools can present a measured statistic, a modelled estimate, a figure from a press release and an unsupported or fabricated claim in the same fluent style. They can also produce inaccurate or invented citations.

“Ask any large language model for a stunting estimate for any district and it will supply one, fluently, instantly, and without provenance,” the authors write.

AI systems may also reproduce existing imbalances in published research, including the dominance of English-language sources and evidence from wealthier countries.

At the same time, the organisations responsible for checking information are facing cuts. The WHO’s 2025 restructuring reduced director posts from 76 to around 34, while up to 30–40% of headquarters staff reportedly faced redundancy.

“Sifters” need authority and resources

The authors argue that checking information cannot be left to technology companies or individual users. AI developers sell fluency and speed, rather than independent verification, while individual users cannot assess every claim at the scale required by a health system.

Public health agencies, universities, national statistical offices, research institutes and international organisations therefore need stronger evidence functions.

These institutions already employ people responsible for assessing the reliability of information, including chief scientists, chief statisticians and guideline review committees. However, Buse and Hawkes say these functions were generally created to certify an organisation’s formal publications, not to oversee the large volume of material now being generated with commercial AI tools.

Professor Kent Buse, Professor of Health Policy at Monash University Malaysia, said:

“Every public health institution employs people whose job is to tell evidence from information – including dis- and mis-information: chief scientists, chief statisticians, guideline committees. Generative AI has made them more important than ever, and at exactly the moment their budgets are being cut.

“Our message to public health agencies is simple: give those functions explicit authority over how the institution uses AI, attach provenance to every figure published, and defend your scientists and statisticians out loud when they are attacked. Silence is itself a decision about what counts as knowledge.”

Calls for transparent sourcing

The authors call for public agencies to establish provenance standards for all information they publish or generate using AI. These standards should make clear whether a figure was directly measured, modelled from other data; or generated by an AI system.

They should also identify uncertainty and provide access to the underlying sources wherever possible.

Buse and Hawkes recommend creating evidence functions that can respond quickly to political developments without abandoning verification. They also call for renewed investment in primary data collection and national statistical capacity.

They argue that no major health survey programme should again depend on a single donor or a limited-term private grant.

The article also highlights the political importance of defending evidence institutions. Scientists and statisticians whose work is independent, transparent and accountable should be supported publicly when they are attacked, the authors say. Failure to respond can itself weaken trust in the evidence those institutions produce.

The authors conclude that the post-truth condition “is not a crisis that will pass”. Public institutions will instead have to operate in an environment where information is abundant, attention is limited and plausible claims can be produced at almost no cost.

Institutions that learn to curate, contest and defend knowledge as a public good will remain trusted a decade from now, they argue.

“The rest will have outsourced the question of what is true to whoever owns the model.”

The information of this article was contributed by Professor Kent Buse, Professor of Health Policy, and Professor Sarah Hawkes, Head of the Department of Global Population Health, at Monash University Malaysia.