OpenAI's o3 AI model helped Boston Children's Hospital identify rare disease diagnoses in 18 children whose conditions had long stumped doctors.

What happened: Researchers at Boston Children's Hospital's Manton Center ran the genomes of 376 patients who lacked diagnoses through OpenAI's o3 Deep Research model, which identified new diagnoses for 18 patients across rare neurodevelopmental diseases, neuromuscular disorders, sudden unexplained death, and early childhood psychosis. The research was announced in the New England Journal of Medicine's AI-focused publication, NEJM AI.

Why it matters: A diagnostic yield of 5% is significant because these genomes had already been analyzed multiple times; finding new answers would normally require days of manual work by human geneticists who are already stretched thin. The AI system can process large volumes of existing but scattered medical data that human analysts have limited time to review, potentially democratizing access to diagnosis for patients globally.

What to watch: The research team emphasized that LLM results still require rigorous human review and that the findings are not a panacea—being diagnosed is only an early step toward treatment. Seven of the identified diagnoses were actually rediscoveries, meaning other treatment teams had found the same answers but had not shared them globally, highlighting the importance of knowledge-sharing once new treatments become available.