U.S. ‘AI Manhattan Project’… Deploying LLMs and AI agents even for nuclear issues [Lee Sang-eun’s Washington Now]

Trump’s ‘Genesis Mission’ announced last November

U.S. Department of Energy unveils plans for 26 strategic areas on the 12th

The U.S. Department of Energy (DOE) on the 12th (local time) unveiled key initiatives under the ‘Genesis Mission’ announced last year, aimed at strengthening national security and solidifying economic leadership by integrating artificial intelligence (AI) across the nation’s science and technology enterprise. The plan envisions using AI across 26 strategic areas—including manufacturing, energy production, basic science advancement and nuclear security—to break through existing technological limits.

The Genesis Mission is a Trump administration initiative centered on the DOE. It includes building the ‘American Science Security Platform (ASSP)’ using the supercomputers and scientific data of its 17 national laboratories, and cooperating with Big Tech such as Amazon, Microsoft (MS), Google and Meta. It has also been described as an ‘AI-era Manhattan Project.’

The materials released that day detail specific strategic tasks by sector. Plans for aggressive use of large language models (LLMs) and AI agents stand out. The focus is on embracing AI’s role proactively to maintain and strengthen U.S. global leadership.

◆ Using AI for scientific discovery and commercialization

The ‘Genesis Mission’ executive order signed by U.S. President Donald Trump last November lays out three broad directives: △building an integrated AI platform to leverage federal scientific datasets △creating AI agents to train science foundation models, test new hypotheses, automate research workflows and accelerate scientific breakthroughs △integrating national R&D resources. Ultimately, the goal is “to solidify America’s technological edge and global strategic leadership.”

The report identifies as its first task using AI to sharply raise productivity in advanced manufacturing. In particular, it argues AI can address the ‘valley of death,’ where scientific discoveries fail to translate into commercialization. It says generative AI and digital twins will be used to identify new manufacturing pathways and strengthen supply-chain resilience.

In science and technology, the report expects AI to serve as a “breakthrough” catalyst. In quantum computing, it aims for AI to tackle the complexity of quantum operations that human researchers struggle to explore, enabling the design and automation of new quantum algorithms. It will also accelerate development of “physics-informed AI” that goes beyond data analysis to understand the principles of nature. In space and particle physics, it sets out an ambition to derive insights into the universe’s fundamental principles through AI models that simultaneously learn from data ranging from the quark-level microscopic world to cosmological observation datasets.

In nuclear power and fusion, the DOE said it plans to use AI to transform reactor design and permitting, cutting construction timelines in half and reducing operating costs by more than 50%. “A significant portion of nuclear plant costs stems from stringent regulatory compliance and paperwork,” the DOE said, adding that “if AI learns safety standards to automatically draft regulatory documents and conduct risk analysis, planning and documentation time can be reduced by more than half.” It also said it would “use explainable AI solutions including surrogate models, agent-based workflows, autonomous laboratories and digital twins, and employ digital twin technology that interprets complex operational data in real time to reduce operating costs.”

To respond to surging demand from data centers and industry, the document says it will modernize the grid by using AI to “accelerate grid decision-making by 20 to 100 times.” It aims to apply deep learning and reinforcement learning to big data sources to reduce uncertainty in grid planning, operations and security, and to build out the grid faster. Beyond making decisions up to “100 times” faster, the report notes it targets “improving power costs and reliability by at least 10%.”

It also includes a plan to secure U.S. industrial leadership in “microelectronics.” The goals include “ensuring sustained U.S. leadership in the global semiconductor landscape, achieving advances beyond Moore’s Law for AI computing and national security applications, and securing global leadership in 6G (sixth-generation) communications networks.”

The DOE also expects AI to dramatically speed up discovery and development in the process of finding and managing strategic materials and subsurface energy resources.

◆ Aggressive use of AI for nuclear security

Notably, the U.S. government also declared it would make bold use of AI in the nuclear security domain. The DOE said, “Deterring America’s adversaries from employing strategic weapons is more urgent and complex than ever,” adding that “to meet these threats, greater flexibility and innovation in nuclear weapons production capacity are required.”

In this context, the DOE said “the transfer of weapons systems between design agencies (Defense Department-related agencies, DAs) and production agencies (Defense Department-affiliated agencies, PAs) is slow and inefficient,” and added it will “develop an AI-accelerated, model-based systems engineering platform that spans and integrates design and production processes.”

The U.S. will begin by turning vast amounts of accumulated data over the past 80 years into usable assets. The report says the U.S. “possesses extensive classified scientific experiments and nuclear weapons test records spanning more than 80 years, as well as unclassified historical nuclear science materials,” noting that “much of this critical information exists only in handwritten notes, printed documents, or photographs,” and stressing it “must be converted into usable data.” The report added that this “aligns with President Trump’s executive order requirement to make federally owned datasets a core component of the Genesis Mission.”

The DOE also explained that in the event of a nuclear accident or radiological threat, it will strengthen public safety and national security by using AI-based fusion systems to cut the time from detection to response “from the current ‘days’ to ‘hours.’” The U.S. said it will deploy AI analytical agents going forward to rapidly trace the leakage routes and origins of nuclear materials, maximizing deterrence against hostile actors.

The document further notes that “not all challenges to nuclear deterrence come from external sources,” and that “regulatory procedures for high-risk facilities are slow and fragmented, creating inefficiencies that can undermine safety and operational efficiency.” It pointed out that “these largely well-intentioned policies are no longer compatible with the urgency of the National Nuclear Security Administration (NNSA)’s modernization and production missions.” The document said it will “introduce auditable, policy-based AI (LLM + agents) that understands safety-based requirements, automates safety analysis and documentation, autonomously configures and runs large-scale simulation campaigns, and continuously develops work plans.” However, it added that “a citable and trustworthy digital regulatory corpus must be built, and these tools must be integrated into a data system with an execution environment (harness) for validation and testing of AI outputs, strong access controls, and end-to-end audit logs.”

◆ Risks barely mentioned

Michael Kratsios, Director of the White House Office of Science and Technology Policy, said in releasing the report that “the United States has doubled the pace of scientific and technological discovery by mobilizing unmatched scientific infrastructure and AI creativity.” He emphasized that “these 26 tasks are a direct call to action for America’s researchers and innovators to join the Genesis Mission and deliver scientific and technological breakthroughs that benefit the American people.”

However, the report largely presented an upbeat view, mostly avoiding discussion of risks associated with integrating sensitive data into AI, the possibility of AI operating beyond control, or the risk of data leakage. It also offered little mention of practical constraints that could become real bottlenecks. Rather than a detailed set of guidelines, it is closer in nature to a forward-looking “declaration” outlining the direction of future policy.

Meanwhile, Japan became the first partner country for the Genesis Mission last month. According to Nihon Keizai Shimbun, Japan’s Ministry of Education, Culture, Sports, Science and Technology signed a document on the 27th at an international conference held in Osaka last month to cooperate with the DOE on the Genesis Mission. The Trump administration has been launching initiatives in succession, including the “Posse Initiative” for cooperation on critical minerals, “Project Vault” to stockpile 60 days’ worth of critical minerals, and “Pax Silica” for supply-chain cooperation in AI and semiconductors. Through these efforts, it aims to form a “U.S.-centered economic bloc” and restore critical supply chains that are not swayed by China and others.

Washington=Correspondent Lee Sang-eun selee@hankyung.com