Conference Agenda

The Nuclear Energy Agency (NEA) and its Committee on the Safety of Nuclear Installations (CSNI) has been pioneering since 1970 in advancing thermal hydraulics and reactor safety assessment through various projects and methodologies, significantly contributing to T/H nuclear safety research. Despite challenges, the progress made provides a strong foundation for future advancements in nuclear safety research and application.

A renewed effort is required to extend thermal hydraulics research beyond traditional water-cooled reactors (WCR) to non-WCR and advanced reactor designs and beyond the traditional end-uses of nuclear. The extensive experience gained from water-cooled nuclear reactors is fundamental in advancing the safety and reliability of next-generation nuclear technologies.

Experimental data are precious, rare, expensive and yet still needed! The first priority is preserving the competences, expertise and database derived from past research investments. And such data needs to be well known and used! It is also fundamental to create new ones in a prompt manner, because “the train was already leaving the station”. Defense in depth is based on knowledge and on the best estimation of uncertainties in risk assessment.

International cooperation is fundamental for identifying and advancing future nuclear safety research and applications and sharing investments. While CSNI constitutes a forum for international cooperation, the scope of activities is limited by resource availability, necessitating the exploration of new, more efficient collaborative models. Thus, to address future challenges and accelerate progress in accident analysis and management, innovative approaches to international collaboration is essential.

Not Submitted

This speech intends to cover four points: independence of regulatory authority, learning from accidents, regulatory challenges, and what we now expect from research.

Innovation requires strong and independent regulatory body. Regulation is often perceived as an obstacle to innovation. Many good operators and vendors may achieve the adequate level of safety even without any regulation, but the failure of a single poor competitor can drive all remaining technologies out of the market, taking an unreasonably long time and effort to recover from. There are numerous examples of conflicts of interest leading to poor decisions by organizations and their leaders. Maintaining the independence of regulatory authorities is essential for demonstrating and deploying advanced nuclear systems. Any erosion of regulatory independence puts the people and the environment at risk and significantly undermines public trust in nuclear technology.

Many people now seem to be trying to believe that severe accidents can be practically eliminated by design. However, new accident scenarios should be considered for new designs. There is still much to be learned from past accidents. The new design or feature or new practice shall also be adequately tested to the extent practicable before being brought into service, and shall be monitored in service to verify that the behavior of the plant is as expected.

There are numerous regulatory challenges, e.g., defining licensing basis events (LBEs). There are cases where the classification of states, such as normal operation, anticipated operational occurrences (AOOs), design basis events (DBEs), and design extension conditions (DECs), may need to be changed. There are also cases where the concepts of severe core damage or loss of containment function of specific barriers do not adequately describe the respective states.

For research, it is urgently needed to organize a framework to lead collaborative research projects with scaled experimental infrastructures to enhance development, validation and benchmarking of state-of-the-art codes, training and education.

This paper explores how artificial intelligence (AI)—particularly multimodal foundation models (FM), intelligent digital twins (IDT), and LLM-based multi-agent systems—is reshaping nuclear thermal-hydraulics and safety analysis (NTHSA). Historically grounded in physics-based modeling, structured validation, and expert-guided reasoning, NTHSA now faces growing demands for more adaptive, predictive, and transparent methodologies. Emerging AI technologies offer the potential to augment these foundations, enabling a shift from static safety analysis to a dynamic, epistemically intelligent safety paradigm.

The paper introduces foundation models—large-scale AI systems trained on diverse textual, numerical, and visual data—as tools that can reason, generalize, and automate complex tasks such as PIRT generation, closure model selection, and physics-code scripting. When embedded within intelligent digital twins, AI can enable real-time plant monitoring, anomaly diagnosis, and adaptive margin management, all grounded in both operational data and physics-based simulations. The integration of multi-agent architectures further allows the decomposition of safety analysis workflows into autonomous, collaborative AI roles—streamlining V&V, optimizing test matrices, and ensuring traceable, auditable recommendations.

This AI-enhanced framework not only accelerates traditional EMDAP loops but also opens the door to earning back conservatism through evidence-based learning. By dynamically reducing epistemic uncertainty over time, safety margins can be optimized while maintaining robust defense-in-depth. Case studies—such as the NAMAC framework and GPT-based discrepancy checkers—illustrate how AI can act as an assistant or advisor, improving explainability, trust, and operational awareness.

The paper also highlights critical challenges: limited nuclear data, explainability of black-box models, online V&V, cybersecurity, and human factors. It advocates for incremental adoption—starting with pilot deployments in non-critical systems, and expanding under transparent, auditable, and regulator-engaged oversight. Emphasizing ethics, human-AI collaboration, and sociotechnical integration, the paper charts a path toward AI as a trusted partner in nuclear safety.

Ultimately, AI is not portrayed as a silver bullet but as a transformative augmentation to the safety toolkit—empowering engineers and regulators to maintain high standards of performance and safety in an increasingly complex operational landscape.

The rapid advancement of artificial intelligence (AI) is reshaping global energy demand, notably increasing the need for stable, carbon-free power sources. As AI-driven services and data centers expand, major economies are revisiting nuclear power as a reliable energy solution. This paper analyzes recent discussions from NURETH-18 through 20, highlighting AI’s emerging role in nuclear thermal-hydraulics and system diagnostics. It further examines global nuclear expansion trends in response to projected electricity demand growth and decarbonization goals. The development of small modular reactors (SMRs), with enhanced safety and modular construction, is accelerating worldwide. Concurrently, regulatory frameworks are evolving to accommodate advanced reactor technologies, as exemplified by the U.S. 10 CFR Part 53 initiative. AI applications in nuclear operations, including anomaly detection, predictive maintenance, and documentation analysis, offer opportunities for efficiency and safety gains but raise new challenges in verification, regulation, and accountability. This paper addresses both technical and regulatory challenges for developers and regulators in adopting AI and deploying next-generation reactors. It concludes that while AI is driving power demand, it also holds potential to support nuclear innovation—provided appropriate safety, governance, and validation mechanisms are established.

Not submitted