In the beginning, people created computers. Some of them said, “Let there be software,” and it was mostly good. Then they said, “Let the software be more intelligent,” and they called that intelligence artificial. And in November of 2025 the White House launched an AI program called the Genesis Mission.
Last November, by executive order, President Donald Trump tasked the Department of Energy—which oversees the nation’s nuclear stockpile—with leading a “dedicated, coordinated national effort to unleash a new age of AI-accelerated innovation and discovery that can solve the most challenging problems of this century.”
Put more simply, the mission aims to build an AI platform, in partnership with universities and private companies, to tackle research in areas ranging from advanced manufacturing to biotechnology, nuclear energy to quantum information science, critical minerals to semiconductors.
On supporting science journalism
If you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.
Most public descriptions of Genesis emphasize its scientific mission, much as the DOE itself tends to foreground supernova research and disease modeling over the warheads it maintains. But the 26 challenges the department released present a more martial side to the mission: seven focus on nuclear weapons and national security.
It makes sense that the DOE would be in charge of a big AI-for-research project, says Bahrad Sokhansanj, a research scholar at the Institute for Law and AI in Boston, who used to work at Lawrence Livermore National Laboratory. Because of both its research on nuclear weapons and its existing basic-science portfolio, the agency already has the right infrastructure. “It has labs, it has computers,” Sokhansanj says. “It has a lot of resources that are relevant to the future of science and technology.” (Lawrence Livermore, Los Alamos National Laboratory and DOE headquarters did not respond to requests for comment.)
In fact, some of the world’s most powerful supercomputers live at the DOE labs. Genesis, Sokhansanj continues, is a way to focus those resources on a single strategic objective rather than letting them remain scattered across the government’s usual silos.
Still, folding AI more deeply into nuclear work can sound scary. Herbert Lin, a senior research scholar at Stanford University’s Center for International Security and Cooperation, sees it less as a giant leap than as the next step in a long computational progression. “AI is just another way of doing computers,” Lin says. “And computers can be applied to everything. So that means AI can be applied to everything in weapons.”
Genesis’s nuclear-AI plans live somewhere between bad grammar and the launch codes.
People hear “AI plus nukes,” Lin says, and jump straight to the most cinematic scenario: an autonomous system deciding to launch an atomic attack. Fortunately, nobody’s actually proposing that. No one in the military, the National Security Council or the acquisitions world wants that arrangement. “Even the president, who is known for saying very nutty things, has not said that,” Lin says.
The idea of putting AI in charge of nuclear launches sits at one end of what Lin deems the stupid-benign spectrum. At the other end is something like using an AI grammar tool on a nuclear-policy memo. “There’s a whole range of stuff in between,” he says. Genesis’s nuclear-AI plans live somewhere between bad grammar and the launch codes.
Several of those challenges aim outward, at protecting people from nuclear threats. For instance, Genesis will use AI to combine data from sensors, simulations and intelligence to detect and respond to potential nuclear or radiological attacks. Another challenge will speed up nuclear forensics, helping analysts to characterize radioactive material recovered after an incident or from a weapon and trace where it came from and whom to hold accountable.
Genesis will also take on nonproliferation, using AI to analyze “satellite imagery, sensing, open-source and government data” to find anomalies that suggest dangerous radioactive stuff might be making its way somewhere it shouldn’t be. It will also reach inward, into the work of maintaining and modernizing the nation’s arsenal. The DOE’s stockpile-stewardship program seeks to understand how aging nukes behave and how modernization efforts affect safety and performance.
To help with that, Genesis will use AI to trawl the DOE’s archive of old data: the results from classified experiments and weapons tests, along with unclassified nuclear research sometimes still trapped in hand-scrawled notes, paper files or film photographs. “No doubt there’s interesting stuff in there they haven’t found,” Lin says. The system would also ingest and process new data.
At least for now those new data will not come from tests of nuclear weapons. But the DOE does study weapon behavior at places such as Los Alamos’s Dual-Axis Radiographic Hydrodynamic Test facility, which uses giant x-ray machines to make high-speed movies of imploding mock materials. One Genesis challenge will develop an automated system that will act like an admin, helping to schedule experiments, steer them in real time and run live diagnostics.
At high-hazard facilities where weapons modernization and production take place, Genesis would also use AI in safety planning for what outline documents describe as “streamlining production” and “removing red tape” (and not the red tape that’s there for good reason). Closer to the scarier end of Lin’s spectrum is a challenge meant to tighten the handoff between the groups that design nuclear weapons and those that produce them, so modernized systems can move ahead faster.
Those seven are the most direct nuclear-security challenges. But the rest of the 26 sit closer to national security than they first appear, Sokhansanj says. The U.S. wants to “win” at AI and also wants self-sufficiency in areas such as critical minerals and manufacturing to reduce its dependence on other countries. Scientific leadership has been a form of American soft power since the end of World War II. In the DOE’s world, though, even the soft stuff tends to have hard-power implications.
A lot of the science the DOE wants to accelerate is dual use, Sokhansanj says. “If you’re working on particle accelerators, if you’re working on advanced factories, that’s all going to have other applications.” A particle accelerator, for instance, can also help reveal how particles inside a weapon behave.
Given that, Sokhansanj says, Genesis brings up potential governance issues. If the DOE is speeding up science, the results will not just be cool; they also could be dangerous—synthetic biology research that reveals, for instance, roadmaps to bioweapons. Sokhansanj suggests the DOE should use some of Genesis’s resources to fund countermeasures and cybersecurity so science secrets and breakthroughs aren’t hacked out into the world.
Lin has a related warning for this Venn diagram overlap of AI and nuclear—particularly when evaluating actual or potential attacks. “There are a lot of people who think that AI is an oracle that can tell you truth,” he says. “And it can’t.”
