On May 29, OpenAI took a paradoxical step. Its life sciences model GPT-Rosalind, released in April to help drug companies run experiments by tapping 50 research databases, is now being given away free to governments and a group of "trusted developers" specifically for biodefense purposes—with OpenAI footing the bill.
The irony? Over the past two years, OpenAI and Anthropic have been among the loudest voices warning that AI could be used to create biological weapons. Now OpenAI is actively distributing a model that can accelerate biological experiment design.
The program is called Rosalind Biodefense
OpenAI has split it into two tracks, collectively termed "defensive acceleration":
- Developer track: Academic institutions, nonprofits, government-affiliated teams, and small-to-medium teams with clear public-interest goals can apply. After review, OpenAI covers the usage fees. Use is limited to defensive work such as pandemic preparedness, early detection, screening, and vaccines.
- Government track: Specific departments of the U.S. government and allied nations get access for early warning, outbreak response planning, diagnostics, and medical countermeasure development.
Partners already announced include Lawrence Livermore National Laboratory, Johns Hopkins Applied Physics Laboratory, the Coalition for Epidemic Preparedness Innovations (CEPI), and biosecurity firms Fourth Eon and SecureDNA. OpenAI said it briefed the White House and several federal agencies in advance.
To demonstrate the model's utility, OpenAI noted that a previous collaboration between GPT-5 and Ginkgo Bioworks cut the cost of cell-free protein synthesis by roughly 40%.
Why defenders get it first
This is the core logic. OpenAI's announcement included a telling line: "Frontier AI should meaningfully advantage those defenders."
The reasoning: AI-accelerated biological research is a double-edged sword. The same model that helps design a vaccine could theoretically help design something else. That risk cannot be blocked—the capability exists. OpenAI's bet is to arm the vaccine makers, diagnostic developers, and early-warning systems first, so defense outpaces offense.
This differs from simply withholding the model. If OpenAI keeps it locked away, the capability still exists and others will build it. By actively arming defenders, OpenAI at least ensures the first-mover advantage stays on the side it wants.
Background: Money is flowing in this direction
The timing is no coincidence. The U.S. Department of Defense has a $200 million AI pilot this year; the fiscal 2027 defense budget request has risen to $1.5 trillion, up 44%; and the Department of Energy has approved $1.2 billion in AI funding. AI's integration into national security is visibly accelerating.
OpenAI's move is half technical, half positioning. By placing itself as a partner in national biosecurity, it gains a higher standing than simply selling APIs.
Whether the free model ultimately prevents the next pandemic or opens a convenient door for certain capabilities is a question OpenAI itself probably cannot answer with full certainty.
Sources: CocoLoop, OpenAI launches biodefense program (Axios); OpenAI is giving away its life sciences AI model to help governments prepare for the next pandemic (The Decoder); OpenAI launches Rosalind Biodefense, offers federal agencies early access to its life-sciences model (R&D World)