OpenAI Foundation pays to create high-quality biological data to train medical AI

The nonprofit foundation that controls OpenAI, the OpenAI Foundation, has announced a new program called 'Public Data for Health', whose goal is to fund the creation of 'high-quality scientific datasets' that would allow AI models to make major leaps in medicine. The underlying idea, as the foundation itself has explained, is that artificial intelligence will not be able to deliver meaningful advances against disease unless researchers can feed models far more information than is available today. Morgan Levine, former vice president of computation at the longevity company Altos Labs, sums it up this way: 'everyone recognizes that data is the biggest bottleneck to successfully applying AI to biology'.

One of the most striking projects funded in this first round comes from an idea by Ruxandra Teslo, a policy analyst specializing in clinical trials. Last year Teslo proposed using data from bankrupt biotech companies: by bidding in their bankruptcy proceedings, it would be possible to obtain detailed regulatory filings, manufacturing strategies and safety data that are normally treated as trade secrets. She herself dubbed these documents 'biotech's lost archive' and suggested they could be used to train AIs capable of acting as copilots in the often opaque drug approval process. The proposal has received a $500,000 grant and will be developed by 1Day Sooner, an organization that represents clinical trial volunteers and which Teslo advises.

In addition to this project, the initial round of grants includes $40 million for a program collecting data on new cancer vaccines at the University of North Carolina at Chapel Hill, as well as support for OpenAdmet, an initiative that runs competitions in which researchers try to predict drug effects. In the foundation's own words: 'we expect that many of the remaining advances in preventing and curing disease will come from combining the intelligence of new models with more observations of the world; in other words, more data'.

The financial backdrop to this initiative is notable. OpenAI began as a nonprofit, but its leader, Sam Altman, restructured it to create a for-profit corporation that develops models, launches products and is now planning an IPO that could value the company at one trillion dollars. Because the foundation holds a 26% stake in OpenAI, it is on track to become the richest charity on the planet, with a stock value that could reach $250 billion; by comparison, the Gates Foundation and an associated trust together held about $180 billion at the end of 2025. Even so, putting that money to work will not be easy: the San Francisco-based foundation is still hiring for key positions and has only begun to accelerate its giving this year. Its largest single donation to date, $100 million, was awarded in August to the Common Health Coalition, an organization that helps patients access hepatitis C drugs. Jacob Trefethen, an executive at the foundation, notes that it operates essentially independently of OpenAI, though it shares the official mission of ensuring that AI 'benefits all of humanity'; according to him, the foundation expects to give away $1 billion before the year is out.

This philanthropic effort comes at a moment when apocalyptic fears have surged about the risk that an out-of-control AI could end human life, possibly by releasing a lethal biological weapon. Those fears have been fueled by insiders at the AI companies themselves, some of whom put the probability of human extinction within the next decade at 10% or more. Last week, both Altman and Elon Musk backed a call by Anthropic chief executive Dario Amodei to 'slow the pace at which we improve the capabilities of AI models' to give risk prevention time to catch up.

As for the 1Day Sooner project, its president and co-founder, Josh Morrison, explains that the $500,000 grant will be used to demonstrate that the organization can obtain these archives from bankrupt companies, which he estimates could be acquired as non-exclusive copies for 'a few tens of thousands of dollars' each. The organization currently holds three such datasets, two of them donated by Lumen Bioscience, a biotech that had previously used the same Chapter 11 strategy to obtain drug development information from another company. Morrison acknowledges that two other attempts to secure pharmaceutical company archives this year fell through because 1Day Sooner's bids were not accepted.

The article frames these bankruptcies as part of what some are already calling a 'new gold rush' for AI training: last month, Google won a bid for the corporate data of bankrupt carrier Spirit Airlines, including 100 million emails, which drew objections from flight attendants and others affected, concerned about the exposure of private or proprietary information.

The pharmaceutical archives 1Day Sooner is pursuing are known as 'common technical documents' and typically contain all the exchanges between companies and regulators, along with detailed scientific and medical measurements; in essence, they capture everything known about a drug. According to Teslo, a contributor to the publication Works in Progress and a nonresident fellow at the Washington-based think tank Institute for Progress, an archive of this kind could turn an AI into a regulatory expert, which in her view would be one of the main ways AI could speed cures to market. In her own words: 'people say we'll invent AI and AI will cure cancer, but that's very far from reality, which is much more complicated, and from the regulatory process'. She adds that 'about 70% of the money and time in drug development goes into clinical development — organizing the trials and testing the drug — but despite that the process is basically a black box, especially for the small biotechs that generate the innovations'.