Federal Health Agency Bets $98.5 Million on AI to End the Rare Disease Diagnostic Odyssey

Imagine spending seven years and visiting a dozen specialists just to get a name for what's wrong with your child. That's not a hypothetical. It's the reality for a huge share of the roughly 30 million Americans living with a rare disease, and it's the problem a new federal initiative is trying to solve.

The Advanced Research Projects Agency for Health, the innovation arm of the Department of Health and Human Services, announced four contracts this week worth up to $98.5 million combined over nearly five years. The program, called RAPID (Rare Disease AI/ML for Precision Integrated Diagnostics), aims to use artificial intelligence to close the gap between when symptoms first appear and when patients actually get answers.

The stakes here are not abstract. Rare disease patients often bounce between doctors for years, racking up medical bills and losing time that matters most when a treatment could actually help. Think of it like trying to find a single mislabeled book in a library with no catalog system, scattered across a hundred different branches that don't talk to each other. That's roughly what rare disease data looks like today: genetic test results sitting in one system, clinical notes in another, patient-reported symptoms nowhere at all. RAPID is essentially trying to build the catalog.

"Rare diseases represent one of the greatest unmet needs in medicine, and one of the most important frontiers for AI," said Scott Gorman, the RAPID program manager, in ARPA-H's announcement. He added that building AI-ready datasets and tools designed to learn from sparse, complex data could improve outcomes for patients while pushing precision medicine forward more broadly.

Who's Building What

Four teams will do the heavy lifting, each tackling a different piece of the puzzle.

The University of North Carolina will construct what's described as the largest real-world rare disease dataset ever assembled, built as a privacy-preserving foundation for future research. UNC isn't going it alone. A wide coalition of academic institutions, health data firms, and patient advocacy groups will help integrate clinical records, genomic information, and patient-reported data into one usable resource.

Sage Bionetworks, a data company, will build a secure platform for evaluating how well different AI tools actually perform, using common standards so researchers can compare results fairly. That kind of shared benchmarking matters because without it, claims about AI accuracy in rare disease detection are hard to verify or trust.

FDNA, which specializes in AI imaging analytics, will develop tools for clinical and direct-to-patient use that collect data at national scale, including photos, videos, voice recordings, and patient-reported concerns. Probably Genetic, an AI research platform vendor, will generate synthetic datasets, essentially artificial but realistic stand-ins for real patient data, to support research without compromising privacy. The company will also build patient-facing tools that pull in photos, wearable data, functional assessments, and genomic information, then connect patients to confirmatory testing and clinical trials.

Lukas Lange, Probably Genetic's CEO, called the effort potentially transformative. In a statement about his company's $10 million RAPID contract, he said government funding has historically catalyzed some of the world's biggest breakthroughs, and that RAPID could do the same for precision medicine. He framed the 2026 to 2030 window as a period that could reshape genetic disease innovation "for 400 million people," a nod to the global scale of rare disease burden.

The program has recruited serious backup. Global Genes and the National Organization for Rare Disorders will help ensure patient voices actually shape how the tools get built, not just how they get marketed. Lawrence Berkeley National Laboratory, the NIH's All of Us Center for Linkage and Acquisition of Data, and the Monarch Initiative are also lending support. Perhaps most notably, AI heavyweights including OpenAI, Anthropic, Amazon Web Services, and Google have pledged in-kind resources, from computing credits for large language models to engineering expertise.

There's also a public-facing piece. Researchers outside the four core teams will be able to access RAPID's data infrastructure through Rare Challenges, an open competition platform supported by NASA's Center of Excellence for Collaborative Innovation, which will benchmark emerging AI approaches across diagnosis and disease mechanisms.

None of this happens in a vacuum. For over 15 years, rare disease patients have turned to internet forums and support groups for answers their doctors couldn't provide, a workaround born of necessity rather than choice. And AI's track record on rare disease detection so far has been uneven. Dr. David Klimstra, cofounder of the computational pathology company Paige, told Healthcare IT News in 2024 that progress has been slowed by sluggish adoption of digital pathology platforms, the tools that let AI actually analyze tissue samples in the first place. High digitization costs, logistical headaches, and a century of habit built around glass slides and microscopes have all played a role in that slow rollout.

Why It Matters

The promise here isn't just faster diagnosis, though that alone would be significant for families who've spent years in limbo. Better data infrastructure could also speed up clinical trials by helping researchers identify the right patients faster, and it could lower healthcare costs by catching conditions earlier, before they progress to more expensive, harder-to-treat stages.

But promise and delivery are different things. Building a genuinely representative dataset means including patients across different backgrounds, geographies, and health systems, not just those with easy access to major research hospitals. Privacy protections and consent management will need to hold up under real-world pressure, not just in pilot programs. And the involvement of major AI companies raises fair questions about how patient data gets used, who benefits commercially, and whether safeguards keep pace with ambition.

Still, as Gorman put it, "for millions of patients and families, living with a rare disease still means years without an accurate diagnosis and too few paths to effective treatment." If RAPID delivers even part of what it promises, that timeline could shrink meaningfully, and for people who've spent years searching for answers, that would count as real progress.