Understanding AI for Nuclear Organizations (HOP)

Large language models (LLMs) offer utilities new capabilities to analyze safety-related regulatory documents with fewer resources than traditional methods. The introduction of Artificial Intelligence (AI) into safety-related analysis demands a structured framework to ensure compliance with established verification and validation (V&V) practices. The approach should consider Human and Organizational Performance (HOP) principles to ensure users have a fundamental understanding of the tools, along with a safety conscious mindset. This paper proposes a three-part framework: train personnel through the Systematic Approach to Training (SAT), disclose general usage to internal and external stakeholders and apply rigorous V&V procedures to specialized AI agents. The approach outlined in the paper is intended to align with established standards from the United States Nuclear Regulatory Commission (NRC), International Atomic Energy Agency (IAEA) and Institute of Electrical and Electronics Engineers (IEEE). Utilities should consider such frameworks to design error tolerant systems aimed at preventing analytical errors and demonstrating regulatory readiness.