Last May, a representative from OpenAI arrived at Los Alamos National Laboratory in New Mexico carrying locked metal briefcases. Inside them were the “model weights” — the parameters that define how an AI system thinks — for ChatGPT’s o3 reasoning model. The briefcases were handed over, the model was installed on one of the most powerful supercomputers in the world, and the birthplace of the atomic bomb became one of the most unusual AI deployments in history.
Los Alamos is where Robert Oppenheimer, Richard Feynman, and hundreds of other scientists spent World War II building a weapon that changed warfare forever. It is where Little Boy — the bomb dropped on Hiroshima — was assembled in a building that still stands as a National Parks monument, where visiting requires a safety briefing on radiation and nearby live explosives testing. It is also, as of 2026, where scientists sit at their desktops and chat with ChatGPT about classified nuclear weapons data, more or less the same way the rest of us use it to draft emails or settle arguments about movie release dates.
From Human Computers to a Supercomputer Running AI
The parallel is worth sitting with for a moment. When the Manhattan Project began in the early 1940s, the word “computer” at Los Alamos referred to a person — many of them the wives of scientists on site, performing thousands of equations by hand on bulky analog desk calculators. The work of computation was physically exhausting, painstaking, and human. The first electronic computers arrived at the lab in the late 1940s. Now Venado — the supercomputer that hosts ChatGPT’s model weights — is capable of around 10 exaflops of computing, roughly 10 quintillion calculations per second, powered by 3,480 Nvidia superchips. Jensen Huang’s signature is on one of the cabinets.
As of August 2025, Venado was placed on a classified network. ChatGPT now has access to some of the country’s most sensitive scientific data on nuclear weapons. Scientists at Los Alamos, Lawrence Livermore, and Sandia National Labs can all access it as a shared resource. The interface, according to the Vox reporters who visited the lab, is essentially identical to the ChatGPT the rest of us use — a chat window, a prompt, a response.
What It’s Actually Being Used For
The program is called the Genesis Mission, a $320 million Department of Energy initiative aimed at doubling the productivity and impact of American science and engineering within a decade. At Los Alamos specifically, researchers are using AI to accelerate work on materials science, plutonium aging, disease treatment, DNA research, energy grid resilience, and astrophysics — alongside the nuclear security applications that drew the most attention when the partnership was announced.
The stated goal is acceleration, not automation. No one at Los Alamos is suggesting that ChatGPT will make decisions about nuclear security. The argument, from both the lab and OpenAI, is that AI can compress the timeline between scientific question and scientific answer — that work which once took years might now take months. Given that Los Alamos is simultaneously running America’s nuclear modernization program, which carries a price tag of $1.7 trillion, the acceleration argument is not a small one.
The Comparison That Keeps Coming Up
Scientists and officials at Los Alamos are well aware of the historical resonance. The people who built the atomic bomb are now among the primary users of the technology that many observers — critics and proponents alike — have compared to nuclear weapons in terms of transformative potential and existential risk. That comparison gets made constantly in AI circles, and Los Alamos is one of the few places in the world where it lands with full biographical weight.
What the Vox reporters found on their visit was notably lacking in the doomsday energy that pervades AI conversations elsewhere. The scientists they spoke with were focused, pragmatic, and largely bullish on what the technology could help them accomplish. The lab has a long institutional memory of being at the frontier of technologies that changed everything — and a long institutional habit of proceeding anyway.
The Same Briefcases, a Different Kind of Weapon
There is something almost cinematic about the image of those locked metal briefcases arriving at the mesa where Oppenheimer worked. Los Alamos has always been where America sends its most consequential scientific bets — the place where the question “what if this works?” gets answered under the most controlled, classified, and consequential conditions available. The fact that the latest briefcase contained a chatbot rather than uranium doesn’t make the moment less strange. It might make it stranger.