OpenAI Foundation pays to create biology datasets for medical AI

The OpenAI Foundation has launched Public Data for Health, a grant program that funds open, high-quality scientific datasets so AI systems can better advance medicine across many diseases. In its own announcement, the Foundation said it is supporting more than $125 million in initial grants to nonprofits and universities, arguing that many remaining breakthroughs will come from pairing stronger models with more biological observations.

Why biology needs more public data

In a post by Abhishaike Mahajan and Jacob Trefethen, the Foundation said scientific data are the foundational input to discovery, and that biology differs from fields where AI can contribute without collecting new observations. It expects progress in preventing and treating disease to depend on more measurements of the world, not models alone.

The program is the Foundation’s second science effort in Life Sciences and Curing Diseases, after AI for Alzheimer’s launched in April. Trefethen leads that portfolio. MIT Technology Review reported that researchers increasingly see data, not algorithms, as the main bottleneck when applying AI to biology.

What the first grants buy

The Foundation said initial awards span layers of data from molecules to epidemiology to regulatory knowledge. One example is OpenADMET, which will create open datasets, benchmarks, and blinded competitions to test whether AI can predict how small molecules are absorbed and distributed in the body, a step aimed at making drug development more predictable.

The Foundation cited the common claim that about 90% of drug candidates fail in clinical trials, often because ADMET properties are hard to forecast.

A second project, CTD Commons, will try to preserve and publish regulatory knowledge from failed or shelved drug programs, including Common Technical Documents that usually stay locked away. MIT Technology Review reported that the idea, linked to policy analyst Ruxandra Teslo and pursued by advocacy group 1Day Sooner, received a $500,000 grant to test whether such archives can be obtained and shared.

A third example is the University of North Carolina’s Initiative for Generative Immunotherapy, which will create public multimodal data to improve personalized cancer vaccines. Technology Review said the Foundation would give $40 million to UNC for that cancer-vaccine data effort. UNC’s Alex Rubinsteyn has said personalized vaccines still face gaps that high-quality open data could help close faster.

How this fits the Foundation’s bigger bet

An earlier OpenAI Foundation update from board chair Bret Taylor said the Foundation expects to invest at least $1 billion over the next year across life sciences, jobs and economic impact, AI resilience, and community programs, including early spending toward a previously announced $25 billion commitment on curing diseases and AI resilience. Technology Review reported that Trefethen said the Foundation hopes to give away $1 billion by the end of the year and operates separately from OpenAI while sharing its mission that AGI should benefit humanity.

The Foundation’s strategy note favors connected data that link biological scales, scarce data that cannot be recreated later, and direct measurements closest to clinical reality, while stressing privacy and consent for human data. It invites researchers with ideas to contact science@openaifoundation.org as it iterates.

Whether the first datasets become widely usable training and evaluation resources, and how quickly they change drug and vaccine timelines, is still the open question.

Key facts: Program: Public Data for Health; second Life Sciences effort after AI for Alzheimer’s (April) Initial grants: More than $125M across nonprofits and universities, per OpenAI Foundation Named projects: OpenADMET; CTD Commons (1Day Sooner); UNC Initiative for Generative Immunotherapy Reported sizes: About $40M to UNC cancer-vaccine data; $500K to 1Day Sooner archive pilot (MIT Technology Review)