The U.S. Department of Energy (DOE) has selected Prometheus, an Idaho National Laboratory-led collaboration, to receive a Phase II award of $60 million over three years, subject to congressional appropriations, under DOE’s Genesis Mission. HiddenLayer said this week it has joined the effort, focused on AI security, including identifying vulnerabilities and protecting models and applications against adversarial threats. Prometheus is a 32-partner collaboration spanning Idaho National Laboratory (INL), Oak Ridge, Argonne and Sandia national laboratories, universities and more than 20 industry partners including X-energy, TerraPower and Oklo, which together have raised more than $200 million in industry cost share and another $30 million in industry capital.
Prometheus Targets Reactor Design, Licensing and Fuel Fabrication
The program’s ambitions extend well past research. Prometheus is meant to apply AI across reactor design, licensing, manufacturing, construction and operation, nuclear fuel fabrication and management of decades of legacy nuclear documents, using generative AI, digital twins and agentic workflows alongside accelerated simulation tools, with INL emphasizing that reactor-related applications will keep humans in the loop. When INL announced the collaboration with Nvidia in February, the lab said it was targeting at least a twofold acceleration in nuclear development schedules and operational cost reductions of more than 50%. Those are targets, not results. X-energy, for its part, committed $10 million and access to its proprietary Xe-100 reactor and TRISO-X fuel designs as a technical platform for AI-assisted design and licensing work, a bet that echoes the broader commercial pressure advanced reactor developers are under to prove out faster paths to deployment.
Regulators Are Testing the Same Tools They Will Have to Evaluate
The push is occurring as the Nuclear Regulatory Commission builds its own AI capability. The agency has concluded its existing regulatory framework can generally accommodate AI applications, while flagging areas that may eventually need clarification. The NRC is now piloting its own internal-use generative AI tool to gain efficiencies reviewing a new enrichment facility application, and staff are using the agency’s Microsoft Copilot Chat tool to research regulatory precedents and compare documents during licensing reviews. A related DOE proof-of-concept illustrated how far that can go: a separate AI tool recently generated a 208-page draft license-application chapter from preliminary safety documents in about a day, work that typically takes a team four to six weeks, though a subject-matter reviewer still had to check it. That creates an unusual dynamic in which AI is not only being built for nuclear developers but is appearing inside plant operations too, on both sides of the same regulatory process Prometheus is now trying to speed up.
HiddenLayer’s addition points to a second-order issue that grows more important as those applications advance. Traditional cybersecurity focuses on software, networks and access, while AI systems introduce additional vulnerabilities involving models, training data, AI supply chains and the interactions between models and other systems, risks that carry extra weight when the underlying information includes proprietary reactor designs, fuel data or regulatory documentation. HiddenLayer has not specified which Prometheus applications it will support, how much funding it will receive, or whether its tools will touch safety-related reactor systems, a caution that fits a sector where AI-driven demand is already reshaping how utilities plan nuclear capacity. As Prometheus moves AI into increasingly consequential nuclear workflows, developers will also have to show those systems can be secured and validated without compromising the safety and technical rigor nuclear deployment requires.
