OpenAI launches Rosalind Biodefense, offers federal agencies early access to its life-sciences model

San Francisco-based OpenAI is launching Rosalind Biodefense, a program that sponsors vetted outside developers’ access to GPT-Rosalind, its gated life-sciences model, so they can build pandemic-preparedness and biosecurity tools. Alongside it, the company is opening GPT-Rosalind itself to select U.S. government and allied partners running public-health and biodefense missions. On background, OpenAI says it briefed the White House and several federal agencies on the approach and is moving to onboard public-health-focused federal agencies. The step deepens a government footprint the company has built through the year: a $200 million Defense Department pilot, deployments at Los Alamos, Lawrence Livermore, and Sandia, and an AWS GovCloud distribution deal.

OpenAI describes the two tracks as “defensive acceleration,” writing in the post that it believes “frontier AI should meaningfully advantage those defenders.” Under the developer track, OpenAI says it will sponsor access to GPT-Rosalind, its frontier reasoning model for life sciences, and provide launch support to a first set of teams building defensive tools in areas like epidemiological modeling, early detection, screening, and non-pharmaceutical interventions. The government track extends the same access for workflows such as early-warning systems, outbreak-response planning, diagnostics, and medical countermeasure development.

Lawrence Livermore National Laboratory is pairing the model with its supercomputing and simulation work to design and evaluate medical countermeasures. Meanwhile, Shankar Sundaram, who directs the lab’s Bioresilience Incubator, said the collaboration is “examining how advanced AI tools can help scientists interpret complex data and existing knowledge, identify stronger candidates.” Johns Hopkins Applied Physics Laboratory plans to integrate GPT-Rosalind into a protein-engineering platform that screens mutant enzymes for therapeutics, countermeasures, and biothreat characterization. The Coalition for Epidemic Preparedness Innovations is applying it to its 100 Days Mission for faster vaccine development, including against the current Bundibugyo Ebola outbreak in the Democratic Republic of the Congo and Uganda. The WHO declared that outbreak a public health emergency on May 17, and the strain has no licensed vaccine.

The GPT-Rosalind expansion arrives a little over a week after President Trump postponed an executive order, hours before he was set to sign it, that would have created a government process for reviewing the most powerful AI models before release. With that process shelved, OpenAI is setting its own terms for early government access, in biology, a week after Washington declined to standardize them for everyone.

Trump’s fiscal 2027 budget request, released in April, asks Congress for $1.5 trillion in defense, which the budget calls “a 44-percent increase.” It also stakes $1.2 billion at the Energy Department on AI, enough for seven supercomputers at Argonne and Oak Ridge. Biodefense shows up mostly in the cuts. The same request reduces HHS by 12.5 percent and “refocuses” the Administration for Strategic Preparedness and Response, trimming it by $356 million and eliminating the Hospital Preparedness Program, while sparing medical-countermeasure work and the Strategic National Stockpile.

OpenAI has published a Preparedness Framework, under which it classified July 2025’s computer-use-focused ChatGPT Agent as high-capability in biology and shipped it with safeguards the company says minimize the risk of harm.

In related news, in February, OpenAI and Ginkgo Bioworks reported that a GPT-5-driven autonomous lab cut cell-free protein-synthesis costs by about 40%, OpenAI’s most concrete public wet-lab result to date. The figure came from a full system, with the model working alongside lab automation, literature access and human sample prep, which makes it a measure of the whole setup more than of the model alone.