Artificial intelligence has helped diagnose 18 children at a Boston hospital whose rare illnesses had confounded doctors, in what researchers are calling a “total game changer” for the field of genetic medicine.
A study published in NEJM AI on 18 June 2026 found that OpenAI’s o3 Deep Research reasoning model identified new diagnoses for patients at Boston Children’s Hospital that had previously eluded extensive human analysis. The 18 cases included ten children with rare neurodevelopmental diseases, four with neuromuscular disorders, two who had died suddenly, and two presenting with early psychosis. The results represent an additional diagnostic yield of 4.8 per cent — a figure that, as lead researcher Dr Catherine Brownstein noted, “means an answer for a family” each time.
How the AI model unravelled the mysteries
The research, led by scientists from the Manton Center for Orphan Disease Research at Boston Children’s Hospital, reanalysed 376 genomes from patients whose rare conditions had remained undiagnosed despite years of testing. The Manton Center, established in 2008 and among the first centres globally dedicated solely to rare diseases, routinely screens patients’ complete sets of DNA against newly identified genes in the hope of finding a diagnosis. But the process is painstakingly slow. “A researcher can only spend so much time on a single case,” said Suyash Shringarpure, a researcher at OpenAI who focuses on the health sector and co-authored the study. “Maybe a case remained unsolved when it came to them first, but a year later, a paper was published that clarifies the link between the gene and the disease.”
The AI model was designed to overcome those cognitive limits. Researchers fed it de-identified clinical and genomic data — including doctors’ notes, patient symptoms, and lists of genes that might be responsible — and asked it to generate evidence-linked hypotheses. The model, released in April 2025, synthesised vast amounts of fragmented information: genetic sequences, clinical features, and the latest scientific literature. “We combine genetic information, phenotypic information, literature search, and the reasoning of AI to deliver diagnoses to families that were once left without any answers,” Dr Brownstein said. She is an assistant professor at Harvard Medical School and scientific director of the Manton Center’s Gene Discovery Core.
All of the model’s suggestions were then reviewed by human experts, who performed additional testing and clinical confirmation before a final diagnosis was made. OpenAI explicitly states in its service terms that its technology should not be used for self-diagnosis, and the study emphasised that the AI acted solely as a hypothesis-generating tool, not an independent diagnostician. Notably, seven of the 18 diagnoses turned out to be “rediscoveries” — conditions that had already been documented elsewhere but were not reflected in the records provided to the study, highlighting persistent problems with data integration across healthcare systems.
Boston Children’s Hospital has already embedded AI deeply into its operations: more than a third of employees use AI tools, saving an estimated 60,000 hours and redeploying over $7 million in labour costs. The hospital reports that its internal AI efforts have led to more than 40 previously unsolvable rare disease diagnoses to date. The Manton Center, supported by a grant from the OpenAI Foundation, now aims to develop a platform-agnostic, low-cost AI copilot for rare disease analysis that could be adopted more widely.
A diagnosis after years of uncertainty
Among the patients given an answer by the AI-assisted workflow was Kyra Benton. She began experiencing concerning symptoms at the age of nine: walking on her tiptoes and struggling to run with a normal gait. Her health deteriorated over the following years as doctors failed to identify the root cause. Just before she turned 20, researchers finally diagnosed her with myofibrillar myopathy, a progressive genetic neuromuscular disorder. “Quite frankly, I’m the type of person that’s not all that much in favour of AI,” she told NBC News. “On the other hand, I do acknowledge that it does have its advantages.”
Benton’s case is one of 18 that underscore a broader challenge: rare diseases affect an estimated 30 million people in the United States alone, and the average patient endures what clinicians call a “diagnostic odyssey” that can stretch for years or even decades. Dr Brownstein described the core problem as one of cognitive limits rather than lack of effort, and the study suggests that periodic reanalysis using AI could make expert-led reviews more scalable as medical knowledge evolves. The research was published in NEJM AI, a peer-reviewed journal launched in December 2023 that is dedicated to the application of artificial intelligence in medicine.
The ethical implications remain significant. AI systems require large volumes of sensitive patient data, raising concerns about privacy, security, and algorithmic bias if training data is not representative. In the UK, where initiatives such as the NHS AI Lab and government funding are accelerating the use of AI in healthcare, the Nuffield Council on Bioethics has highlighted issues including erroneous decisions, accountability, bias, and public trust. Companies such as Mendelian are already using AI to identify rare disease patients within NHS primary care practices, but legal and ethical frameworks are still evolving. For now, the technology’s role at Boston Children’s Hospital — and in the lives of families like Kyra Benton’s — remains that of a powerful assistant, not a replacement for the human clinicians who deliver the final verdict.
