The inaugural workshop of the NEA’s AI Platform for Nuclear Research and Education (AIxpertise), convened online from 15 to 16 October 2025, contributed to the further development of the new joint project dedicated to leveraging artificial intelligence (AI) for the benefit of nuclear energy professionals.
With AIxpertise, the NEA seeks to enhance collaboration among nuclear sector stakeholders in industry, safety, research and academia to equip experts with AI knowledge and instruments to advance their research and train the next generation of nuclear experts.
The workshop brought together 45 organisations from 17 countries, including safety organisations, research institutions, academia, industry and Big Tech companies. The programme featured presentations and discussions on the overall project structure, NEA IT and Data Bank services, the legal framework for the AIxpertise project, and a draft programme of work across its three focus areas: data, AI algorithms’ benchmarking, and education & training.
Participants discussed ongoing initiatives and explored ways to enhance and align them through collaboration within the AIxpertise programme.
Data
The Jožef Stefan Institute (IJS), Slovenia, presented the pulse and depletion experiments being carried out at the IJS TRIGA Mark II research reactor and its plans to develop a new versatile European reactor for neutron irradiation and nuclear research, VERONICA. IJS has offered to share TRIGA reactor data with AIxpertise members.
The Institute for Energy Technology (IFE) of Norway presented a review of the Halden Reactor Project (HRP) legacy database, a key resource encompassing over 60 years of experimental work crucial to advancing research and safety assessments.
The team of the Nuclear Research Data System (NRDS) at Idaho National Laboratory (INL), United States, demonstrated Retrieval-Augmented Generation (RAG) with the potential to significantly improve the accessibility and interpretability of the HRP data content.
Oak Ridge National Laboratory (ORNL), United States, demonstrated optimised machine learning-driven exploitation of existing datasets to improve modelling and simulation (M&S) as well as to validate M&S in domains of sparse validation data like high-assay low-enriched uranium (HALEU) application cases, which are important candidates for future small modular reactor deployment.
The participants discussed the AIxpertise project’s scope directed towards fostering data that meet the FAIR principle (findability, accessibility, interoperability and reusability) as a foundation for AI&ML approaches. They expressed strong support for work on AI-driven tools fostering accessibility and interpretability of data, including the HRP data. There was also strong interest in sharing research reactor data and in operators’ suggestions to share nuclear power plant time series data with AIxpertise members protected by the AIxpertise legal framework.
Benchmarking AI algorithms
The advantages of international co-operation in benchmarking activities were outlined in a presentation on the achievements of the NEA Nuclear Science Committee’s Task Force on Artificial Intelligence and Machine Learning for Scientific Computing in Nuclear Engineering. North Carolina State University (NCSU) expressed support for developing international benchmarks within AIxpertise.
Presentations by Imperial College, United Kingdom, on foundational AI for fluids, solids, particle and radiation modelling methodology, and by the INL, United States, on machine learning for nuclear fuels and materials modelling, demonstrated new opportunities for boosting modelling capabilities and accuracies with AI applications.
Microsoft provided a glimpse into the future of Agentic AI proposing efficiency gains by fully integrating AI agents into human-led workflows, with the potential to speed up nuclear licensing processes.
Participants discussed the proposed AIxpertise project scope, intended to benchmark the output of AI algorithms and to ensure their transparency, reliability and readiness for regulatory review.
Hands-on training and best practices
The Institute of Science Tokyo, Japan, discussed the development of knowledge management systems based on supportive AI agents based on RAG technology and the need for comprehensive validation. North Carolina State University, United States, and the Polytechnical University of Milano, Italy, presented existing education opportunities and expressed support to develop and provide training to AIxpertise project participants. Participants discussed the virtues of different training formats to foster nuclear workforce development and continuous learning.
The workshop concluded with a roadmap to launch the AIxpertise project in the beginning of 2026. The next steps are to refine the AIxpertise programme of work based on further feedback from potential project participants and a review of the legal agreement.
Interested organisations are invited to join the AIxpertise development by contacting AIxpertise@oecd-nea.org. The next AIxpertise workshop is scheduled for 21-22 January 2026.
Learn more on the Aixpertise page.
