The NRA cited as a reason for the request the anticipated increase in workload for safety reviews, driven by plant modifications to comply with new regulatory requirements and reactor replacement projects. By introducing AI into review processes—such as comparing with past review records, conducting cross-references of documents via chat, and assisting in drafting review reports—the NRA expects to streamline the confirmation of applications and supporting documents.
The move also reflects the new Act on the Promotion of Research, Development and Utilization of Artificial Intelligence-Related Technologies, enacted in May this year. The law stipulates that “the government shall promote the active utilization of AI-related technologies by administrative agencies to enhance and improve the efficiency of administrative operations.” With rapid advances in large language models (LLMs) and other AI technologies, the NRA aims to apply them to review operations to improve overall efficiency.
The NRA plans to implement AI use in stages. Starting next fiscal year, it will identify areas of the review process where AI can be effectively applied, based on staff needs and available data, and assess the feasibility of AI-supported reviews. In FY2027–2028, the NRA will design and develop systems for review support based on the findings, and evaluate whether AI outputs exceed required quality levels. By FY2029, it intends to build the system and prepare operational manuals, with the goal of introducing AI into review operations from FY2030 onward.
At the press conference, NRA Chairman YAMANAKA Shinsuke remarked, “I hear that AI is already proving very helpful in improving review efficiency in the United States. We would also like to utilize AI to improve efficiency in Japan.”
Asked by reporters about his concrete expectations for AI technology, Dr. Yamanaka responded, “We envision cases such as training AI to recognize sounds from motors or the tapping of concrete for use in anomaly detection. There are parts we will not know until we actually try, but our hope is to shorten review times while maintaining the same safety standards as today.”
