Who filters what AI produces: two experts call on public health to take on verification as its own task

By Bioengineer.org · October 25, 2026.

Global health experts Kent Buse and Sarah Hawkes, of Monash University Malaysia, argue that public health institutions should make evaluating AI-generated content a core responsibility. They set out the case in a peer-reviewed commentary, “Who sifts? Knowledge as a public good in the age of AI,” published on October 21, 2026, in PLOS Global Public Health.

It helps to be clear about what this is. According to the source itself, it is a commentary or editorial, not an experimental or clinical study. It does not present a screening tool, a dataset or a comparison of how institutions perform. The available material also doesn't tell us what supporting examples, methods or limitations it includes. It is a policy argument. It does not measure how much health misinformation is circulating or which remedy works.

The central idea is in the title: knowledge as a public good. If it is one, someone has to take charge of separating what is reliable from what is not, and the authors point to public health institutions as that “someone.”

Our take: the approach asks the right question but leaves the answer open. AI has driven the cost of producing plausible health content close to zero, while verifying it remains expensive and slow. That asymmetry is the underlying problem, and in the short term it is a real risk: wrong medical advice that looks authoritative and reaches patients and professionals.

The hard part is execution. Saying that institutions “must” filter says nothing about the resources, the legitimacy or the speed they would need to keep up with the volume of content. A public agency reviewing by hand cannot compete with an automated generator. The sensible course is for verification to rely on AI as well, with human oversight and source traceability. That requires funding and public trust, both of which are scarce.

There is also a design risk. Making institutions the sole arbiter of truth can breed distrust if they get things wrong or are seen as biased. Credibility comes from being transparent about how content is evaluated, not from claiming authority.

In the long run, the outlook is optimistic. AI that helps diagnose disease, discover drugs and bring us closer to eradicating illnesses will only deliver on that promise if society can trust what it says. Building the verification infrastructure now is the precondition for making use of that abundance of health knowledge. Buse and Hawkes's commentary doesn't settle the how, but it rightly puts the focus on who is accountable.