The AI initiative of Donald Trump's administration, the Genesis Mission, includes tasks related to the U.S. nuclear weapons complex. Journalists at Scientific American drew attention to this after the country's Department of Energy published a list of focus areas.

The agency presented the Genesis Mission as a program to accelerate scientific discoveries using artificial intelligence. However, among the first 26 tasks there is a block related to nuclear deterrence, weapons production, nonproliferation and nuclear forensics.

What the Genesis Mission is

On November 24, 2025, Trump signed an executive order launching new federal research and development programs in the field of artificial intelligence. The U.S. Department of Energy is to deploy the American Science and Security Platform — an infrastructure to bring together the supercomputers of the national laboratories, secure cloud AI environments, scientific data sets, specialized models and tools for experiments using artificial intelligence.

In the order, the initiative is described as a way to accelerate scientific discoveries, strengthen national security and ensure "energy dominance." The platform is to support model training, simulations, AI agents, workflow automation and the use of federal, academic and private data sets. The Department of Energy was named responsible for implementation, and the project was led by Under Secretary for Science Dario Gil.

On schedule, the agency was to define at least 20 national science and technology tasks within 60 days, conduct an inventory of computing resources within 90 days, select starting data sets and models within 120 days, and demonstrate initial operability of the platform on at least one task within 270 days.

The nuclear block on the task list

On February 12, the Department of Energy published a list of 26 Genesis Mission tasks. They cover energy, science, manufacturing, critical materials, quantum technologies, data centers and national security.

Scientific American compared the authorities' public statements with this list and singled out seven areas related to the nuclear complex and national security. At the same time, the published documents do not imply that AI is being given the authority to make decisions about the use of nuclear weapons.

1. Nuclear threat assessment, readiness and response. The Department of Energy proposes creating a multimodal AI system that will combine sensor data, modeling and intelligence reporting. The goal is to shorten the time from threat detection to response from days to hours.
2. Digitizing historical nuclear data. This concerns an archive of classified experiments and nuclear weapons tests, as well as unclassified nuclear science materials spanning more than eight decades. To convert paper reports, photographs and drawings into data sets suitable for search and modeling, they plan to use text recognition, computer vision, information extraction and 3D reconstruction.
3. Increasing the throughput of research nuclear facilities. The agency describes an AI "facility operating system" that will help plan experiments, adjust them in real time and combine diagnostic data with modeling.
4. Integration of design and manufacturing for nuclear deterrence. The description speaks of shortening the cycle between design and manufacturing units in order to deliver new weapons systems "at a fraction of the traditional time and cost."
5. Nonproliferation. The Department of Energy wants to use multimodal models and analytical agents for the joint analysis of satellite imagery, sensor data, open sources and government data. The system is to identify anomalies associated with the possible spread of nuclear materials.
6. Optimizing the operation of high-hazard facilities. The agency proposes using verifiable AI systems based on large language models and agents to automate the analysis of safety requirements, the preparation of documentation and the planning of work. The goal is to reduce planning and paperwork time by more than 50%.
7. Nuclear forensics. The Department of Energy plans to apply multimodal AI for the accelerated analysis of samples and debris, the determination of their origin and the reconstruction of a material's probable history after an incident.

The list also includes a separate task to accelerate the development, production and certification of materials for strategic deterrence. It mentions AI agents for finding and testing new methods of plutonium purification.

AI does not get the "button"

Scientific American emphasized that the scenarios under discussion do not mean handing over to AI decisions about a theoretical nuclear strike. The publication cites the position of Herbert Lin, a senior research scholar at the Stanford Center for International Security and Cooperation: no one in the military, the National Security Council or the procurement system is proposing to delegate launch to an autonomous system.

In February 2026, a Pentagon representative also told The Washington Post that the decision to use nuclear weapons remains with a human, and handing such a function over to AI is not being considered. Experts' concerns are linked to the broader introduction of AI into threat analysis, production, planning, experiments and risk assessment. Lin warned that AI should not be perceived as a source of guaranteed truth, especially when assessing real or potential attacks.

Bahrad Sohansanj, a researcher at the Institute for Law and AI, noted that the Department of Energy is suited for such a program, since it already has national laboratories, supercomputers and data. But he also pointed to governance risks: accelerating science could affect not only beneficial areas but also potentially dangerous ones, including biological and defense research.

Earlier it became known that the U.S. national laboratories would apply new OpenAI AI models for scientific research and for ensuring the security of nuclear weapons.

Recall that in May the Pentagon concluded agreements with Nvidia, Microsoft, Reflection and Amazon Web Services to apply AI tools in classified military environments.

Later the U.S. Army convened leading contractors for deeper integration of artificial intelligence into military systems while maintaining human control.

Source: ForkLog