A 2026 WHO discussion highlights the growing challenge of using AI in health care. While AI helps researchers find and organize information quickly, experts worry about potential errors in evidence. The core issue is trust, as AI-assisted mistakes could eventually influence clinical guidelines and impact patient safety across systems.
A doctor asks an AI system a question. Within seconds, it produces an answer that looks authoritative, cites research and sounds convincing.
The difficult question comes next: should anyone act on it?
That question may matter more for the future of AI in health care than how sophisticated the underlying technology becomes.
Much of the discussion about AI in health care is still about capability. Can it identify a disease? Read an image? Analyse a patient record? Search thousands of research papers? Discover a drug? For many of these questions, the answer is increasingly yes.
But capability is not evidence.
Health care depends on a chain of trust. Research has to be conducted well. Its findings have to be scrutinised and published. Evidence from different studies has to be brought together and understood in context. At some point, that evidence informs a clinical guideline, a policy decision or a clinician's judgement.
AI is increasingly becoming part of how research is done. It can help researchers find and organise literature, analyse information, explore themes and summarise findings. Used well, these applications can make research faster and potentially more accessible.
But research does not end with the manuscript. Once published, findings enter a much larger evidence ecosystem. They may be scrutinised alongside other studies, incorporated into evidence syntheses and eventually inform clinical guidelines or health policy. What begins as a research finding can travel a long way before it reaches a patient.
That makes a different question important: What happens when an AI-assisted mistake becomes part of the evidence on which someone else relies?
A missed study, a distorted finding, a reference that does not exist, or a summary that removes an important limitation may seem like relatively small errors. They become more consequential when that information is subsequently used to build another piece of research, inform an evidence review or support a policy decision.
AI could also strengthen this system. It can help researchers search much larger bodies of literature, identify connections across studies and support faster evidence synthesis. But as AI moves from helping us find evidence to interpreting and synthesising it, a different set of questions emerges. What has it found? What might it have missed? Has it preserved the context and limitations of the original research? And can we be confident that the synthesis reflects the evidence rather than simply producing a plausible answer?
A 2026 WHO discussion paper on AI and evidence-informed health policy highlights both the potential and the risks. It points to AI's ability to accelerate analysis and synthesis, while raising concerns about bias, transparency, equity, data governance and other challenges across the policy cycle. It also emphasises the continuing importance of human judgement and oversight.
This is particularly relevant for India.
India has already begun building part of the necessary foundation. The Indian Council of Medical Research's 2023 ethical guidelines for AI in biomedical research and healthcare provide a framework for ethical decision-making, governance, ethics review and informed consent. ICMR describes the guidelines as a living document, recognising how quickly the field is evolving.
But guidelines alone will not be enough.
Researchers will need to question what AI retrieves and generates. Those involved in evidence synthesis will need to understand what may have been missed or distorted. Publishers will need transparency about how AI has been used. Guideline developers and policymakers will need to consider the provenance and quality of AI-assisted evidence. Clinicians will still need to judge how evidence applies to the patient in front of them.
The challenge is that the more deeply AI becomes embedded in knowledge creation, the harder it may become to see where human judgement ends and machine assistance begins. That makes transparency more than a publishing requirement. It becomes part of the evidence itself. If we do not know how information was generated, selected or synthesised, it becomes harder to judge how much confidence we should place in it.
The question, then, is no longer simply, can AI do this?
It is also, what does the evidence tell us?
And perhaps most importantly, can we trust that evidence enough to change practice?
Those questions take us beyond the algorithm. They take us into the quality of research, the integrity of publishing, the way evidence is synthesised and the judgement used to turn evidence into policy and practice.
India has an opportunity to become a major adopter of health care AI. But the larger opportunity is to build an evidence ecosystem in which AI can accelerate the discovery and synthesis of knowledge without weakening the standards by which that knowledge is judged.
Ultimately, the success of AI in health care will not be measured only by what the technology can do.
It will be measured by the quality of the decisions it helps inform, the care those decisions enable and whether patients are better off as a result.