AI models need more data about biology, and OpenAI is paying to create it

Artificial intelligence (AI) models require vast amounts of high-quality data to make meaningful advances in medicine, particularly in areas like drug development and disease treatment. According to Morgan Levine, a former vice president at Altos Labs, a company focused on longevity research, data is the biggest bottleneck preventing AI from achieving breakthroughs in biology. Currently, AI systems lack sufficient information to navigate complex processes such as drug approvals, which are often opaque and involve extensive regulatory filings. The OpenAI Foundation, the nonprofit parent of OpenAI, has acknowledged this gap and launched an initiative called Data for Public Health to fund the creation of new scientific datasets specifically designed to train AI models in medicine. The foundation aims to address this shortage by supporting projects that generate or curate biological and medical data, enabling AI to assist more effectively in healthcare.

The OpenAI Foundation, a nonprofit organization linked to OpenAI, announced it would allocate $40 million to fund two major projects as part of its Data for Public Health initiative. The first project, led by the University of North Carolina at Chapel Hill, will focus on collecting data about novel cancer vaccines. The second project, OpenAdmet, involves competitions where researchers predict the effects of drugs, helping to improve AI models' understanding of pharmacology. Additionally, the foundation granted $500,000 to 1Day Sooner, an advocacy group, to pursue an idea proposed by Ruxandra Teslo. Her proposal suggests acquiring data from failed biotech companies during bankruptcy proceedings, which could include regulatory filings, manufacturing strategies, and safety data—often treated as trade secrets. These datasets, referred to as biotech’s lost archive, could provide AI with critical insights into drug development and regulatory processes.

Ruxandra Teslo’s idea centers on obtaining data from bankrupt biotech companies by bidding at their bankruptcy proceedings. These companies often hold valuable datasets, such as common technical documents, which contain detailed scientific and medical data about drugs, including interactions with regulators and clinical trial results. Such documents are typically inaccessible due to their proprietary nature. Teslo argues that acquiring these datasets could help train AI to become an expert in regulatory processes, potentially speeding up drug approvals. The OpenAI Foundation’s $500,000 grant to 1Day Sooner will support efforts to test this approach. However, initial attempts have faced challenges, as two bids for drug company files were unsuccessful this year. Despite these setbacks, Teslo believes that bankruptcies could become a new land grab for AI training data, similar to how Google recently acquired Spirit Airlines’ corporate data, including 100 million emails, during its bankruptcy.

The OpenAI Foundation, based in San Francisco, holds a 26% equity stake in OpenAI, the for-profit corporation that develops AI models and products. Due to OpenAI’s planned initial public offering (IPO) and potential valuation of $1 trillion, the foundation could become the wealthiest charitable organization in the world, with an estimated $250 billion in stock value. For comparison, the Gates Foundation and its associated trust held approximately $180 billion in assets at the end of 2025. The foundation has only recently begun ramping up its grantmaking efforts and is still hiring key staff. Its largest single donation to date is $100 million, awarded in August to the Common Health Coalition, an organization that helps patients access hepatitis C drugs. The foundation operates separately from OpenAI but shares its mission of ensuring AI benefits humanity.