On June 18, 2026, OpenAI said it had published a study in NEJM AI, conducted with researchers at Boston Children's Hospital and Harvard, in which its o3 reasoning model helped clinicians revisit long-unsolved rare pediatric disease cases and reach new diagnoses.
NEJM AI · OpenAI × Boston Children's Hospital · Harvard
An AI "Co-Pilot Geneticist" Cracks 40+ Rare Disease Cases Families Had Waited Years to Solve
Using o3 Deep Research, clinicians reasoned across fragmented genetic data, phenotypes and vast medical literature to surface diagnoses that were previously out of reach — ending long "diagnostic odysseys" for children with rare disease.
40+
previously unsolved rare diseases newly diagnosed
50+
internal workflows automated
60K+
hours saved across the hospital
$7M+
equivalent labor reallocated to higher-value work
Adoption across the organization
More than one-third of all staff now use AI — built in as an enterprise-wide layer, not a one-off tool.
How o3 Deep Research reaches a diagnosis
Ingest fragmented genetic data + phenotypes
→
Autonomously search hundreds of sources
→
Synthesize literature beyond human limits
→
Surface candidates → physician makes final call
Strengths
Multifaceted, multi-step investigation
Synthesizes scattered data and literature
Faster, citation-backed research drafting
Limits to manage
Outputs unstable with too little or too much prompting
Risk of hallucinated content
Data-privacy governance is critical
"It enabled diagnoses that were previously impossible — giving hope to families."
The AI acts strictly as support; final clinical judgment remains the physician's responsibility. The study is framed as a step toward clinical evidence for AI in rare-disease diagnosis — accounting for both its capabilities and its limits.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…