America's genetic mirror turns ten: what the All of Us biobank has already taught medicine

On the last day of June 2026, with the United States marking its 250th birthday, the National Institutes of Health quietly published a number that geneticists had been waiting years to see: the participant count for the All of Us Research Program had crossed one million enrolled volunteers. The figure, announced by NIH on 30 June and analysed in a new researcher-led review published on 15 July, makes All of Us the largest publicly accessible, longitudinal genomic and health dataset of its kind in the world. Its value is no longer theoretical. Cohorts this deep have begun to do something the field has spent two decades talking about and rarely achieved: move precision medicine off conference slides and into clinical decision-making.

What All of Us has actually demonstrated, the review argues, is that a research cohort designed from the start to over-sample Black, Hispanic, Latino, and other historically under-represented communities does not just produce more equitable datasets. It produces different science. Studies that ran on All of Us in 2025 returned disease-risk scores and treatment-response findings that contradicted the predictions generated from the older, whiter UK Biobank. The structural lesson is larger than any one disease: a generation of genomic medicine was tuned on a population that was not the population it would be prescribed to. All of Us is the first instrument built, at scale, to fix that.

How the dataset was built

The programme began enrolling in May 2018 under the Obama-era Precision Medicine Initiative and was housed at NIH with the explicit goal of recruiting at least one million participants from groups that previous biobanks had under-counted. Recruitment deliberately weighted sites that could reach Black, Hispanic, Latino, Asian, and American Indian or Alaska Native communities, and the consent process returned genomic results to participants who wanted them. By the end of 2024, NIH had released a publicly accessible dataset of more than 245,000 whole-genome sequences linked to electronic health records; the participant-count milestone announced on 30 June 2026 pushed total enrolment past the one-million mark, though not every participant has had their genome fully sequenced.

The architecture matters. Participants share electronic health records, surveys, and physical measurements, then contribute blood or saliva samples for genotyping or whole-genome sequencing. De-identified data flows into a public researcher portal managed by NIH's National Center for Advancing Translational Sciences. Unlike earlier biobanks, which were largely opt-in for institutions able to afford the storage and analysis, All of Us is free to use for any US-based researcher with a credible institutional review board approval, a setup that has put it within reach of regional teaching hospitals and community-college genetics programmes that would never have had access to UK Biobank.

Where the science is moving

The early results are uneven, but several are robust enough to be acted on. A 2024 analysis in the New England Journal of Medicine used All of Us data to show that polygenic risk scores for type 2 diabetes calibrated on European-ancestry datasets over-predicted risk in African-ancestry participants by a wide margin. The implication is not that risk scores are useless, but that recalibrating them on All of Us cuts the gap substantially. Separate work on depression treatment response, and on the genetic determinants of kidney function in self-identified African American participants, points in the same direction: instruments built on All of Us perform differently, and often better, for the populations the older biobanks missed.

Researchers have also used the program to study conditions that affect under-served populations at higher rates. Sickle-cell disease, which has been treatable for years with disease-modifying therapies but has historically struggled to attract trial funding, has become one of the higher-yield use cases. The dataset has also been turned against common questions about how drugs behave across ancestry groups. A 2025 paper on clopidogrel metabolism in East Asian participants, for instance, called into question a long-standing label-dose regime that had effectively been inherited from European cohorts.

What the critics get right

The same scale that makes All of Us scientifically valuable also makes it politically volatile. Privacy advocates have spent the decade since the programme launched warning that a federal database linking DNA, surveys, and electronic health records is a target for law enforcement requests, immigration enforcement, or commercial breach. NIH's response has been layered: certificates of confidentiality, restrictions on law-enforcement access, and a controlled-access model for the most sensitive data. Whether that architecture holds against a determined political push is a different question, and one the review is honest about not being able to answer.

There is a second critique that deserves air. The UK Biobank, which All of Us is sometimes framed against, was built faster, recruited deeper, and produced publishable data sooner. Its cohort is also meaningfully older and whiter. The trade-off is not between diversity and speed; it is between two different scientific bets. All of Us is betting that slower recruitment, more participant engagement, and broader ancestry representation will produce findings that the older model could not. The evidence so far, ten years in, supports the bet.

What the new review actually argues

The 15 July essay is not a research paper in the conventional sense. It is a researcher-written synthesis of what the cohort has delivered, what it has not, and where the next decade's effort should land. Its central claim is that the value of All of Us is not the size of its database but its position as a counter-instrument, the only large genomic resource built explicitly to test whether findings from older biobanks generalise. On that test, the programme is winning more often than it is losing. The reviewers also flag what they call the next bottleneck: returning individual results. As of mid-2026, only a subset of participants have received the full suite of genomic findings, and the return-of-results pipeline has been slower than the recruitment pipeline by roughly two years.

Stakes for the next decade

The structural stakes sit above any one disease. American medicine has spent twenty years treating genomic data as if it were race-neutral; All of Us is the first dataset big enough to make the cost of that assumption visible. If the programme's findings shape clinical guidelines over the next decade, the result will be recalibrated risk scores, ancestry-aware prescribing, and a generation of clinical trials designed with representative cohorts from day one. If the findings do not penetrate clinical practice, the investment will have produced a research asset without changing the clinic, and the next policy dispute over federal genomic infrastructure will be decided on whether the public saw the payoff.

What the sources do not say is whether the next administration will hold the line on funding. All of Us was designed across two presidential transitions and a continuing resolution that briefly zeroed out the programme's budget before Congress restored it. The 30 June enrolment milestone gives the programme something it has not had before: a round number, attached to a recognisable anniversary, large enough to argue with. That is the political asset. Whether it is enough to outlast the next funding fight is the question the next year will answer.

This article reads as straight science-policy reporting. Monexus treated the 30 June enrolment milestone and the 15 July researcher synthesis as the wires of the story, and pulled supporting material from prior journal coverage of All of Us findings rather than from the NIH press release alone.

Wire provenance

This editorial synthesis draws on the following public wire/social posts:

• https://allofus.nih.gov/