The International RegLab Project has issued its first report, detailing findings from its initial cycle, which centered on the secure and transparent incorporation of artificial intelligence (AI) within the nuclear energy sector. The ‘sandboxing’ initiative facilitates collaboration among technologists, operators, and regulators to assess the advancement of new technologies from their conceptual stages to actual deployment.
RegLab #1 specifically analyzed the use of AI for real-time monitoring of data from nuclear power plants to identify inconsistencies in operations. Participants noted potential advantages, including increased safety margins, earlier detection of anomalies, and decreased operational expenses. Two key challenges were identified: the necessity for explainable AI and reliable data assurance.
The report emphasizes that AI explainability must be supported by robust defence-in-depth strategies and verifiable justifications for applications where safety is critical. High-quality, well-managed, and representative datasets are crucial for establishing credible safety cases that rely on AI. The RegLab approach was commended for promoting productive discussions among various stakeholders.
Recommendations for future activities involve creating working groups to formulate best-practice guidelines for an AI nuclear assurance framework. This framework would encompass AI verification and validation standards, practical boundaries for application, management of residual risks, enhanced training programs, and the standardization of metadata structures.
This collaborative project is conducted under the Nuclear Energy Agency (NEA), in partnership with the Electric Power Research Institute (EPRI) and the International Atomic Energy Agency’s (IAEA) ISOP initiative, with support from several international regulatory bodies.
