OpenAI kicks off a 1,000-scientist AI jam session

What happens when you give over 1,000 elite scientists the smartest AI models on the planet for a day?

The U.S. Department of Energy (DOE) just launched its first-ever 1,000 Scientist AI Jam Session, bringing together researchers across nine prestigious national labs—including Oak Ridge, Berkeley, Argonne, and Los Alamos.

Even better, both OpenAI and Anthropic were brought in to let scientists stress-test their best reasoning engines against real-world research problems.

Frontier reasoning meets the Manhattan Project legacy

Scientists got access to frontier systems built specifically for deep reasoning. OpenAI brought its o3-mini model, while Anthropic deployed Claude 3.7 Sonnet, its newly released hybrid reasoning model.

Instead of treating AI like a glorified search engine, researchers put the models through the entire scientific loop: literature review, hypothesis generation, experiment planning, code writing, and raw data analysis.

Energy Secretary Chris Wright and OpenAI co-founder Greg Brockman visited Oak Ridge National Laboratory to rally the researchers. Wright did not hold back on the rhetoric, comparing the public-private effort directly to the Manhattan Project and framing AI leadership as a national security race America must win.

From classified bunkers to everyday labs

Both AI labs have been steadily expanding their footprints inside government defense and research circles.

OpenAI signed an agreement last month to deploy o-series models across national labs for astrophysics, clean energy, and materials science, while testing multimodal models at Los Alamos for biosecurity safety.

Anthropic had already laid significant groundwork. Back in April 2024, it became the first frontier AI developer to evaluate a model inside a Top Secret classified environment with the National Nuclear Security Administration to probe nuclear risks.

The findings from this massive jam session won't stay behind closed doors either. The teams plan to publish a follow-up report outlining exactly where frontier models succeed and where they fail in hardcore scientific inquiry.

Why it matters

Science is the ultimate proving ground for frontier AI, because scientific inquiry tolerates zero hallucinations and demands rigorous, multi-step logic.

For researchers and lab teams, this shift marks the graduation of AI from a basic writing tool to a genuine co-investigator. Models capable of extended reasoning can dramatically shorten the time needed to map chemical compounds or simulate clean energy experiments.

From an industry standpoint, we think OpenAI and Anthropic are executing a brilliant play for state-level legitimacy. Putting models directly into national labs gives them unmatched, high-density feedback from domain experts that standard consumer benchmarks cannot replicate.

The catch? As government leaders increasingly view AI through a Manhattan Project lens, the line between open scientific innovation and geopolitical defense tech is practically gone. That alignment opens massive funding spigots today, but it could severely restrict how openly frontier models are shared tomorrow.