Safety, risk, and reliability analyses are integral to the design, development, and operation of nuclear reactors, ensuring compliance with regulatory requirements for licensing and operational safety. These analyses, culminating in the Safety Analysis Report (SAR), demand extensive time, resources, and multidisciplinary expertise, making them one of the most labor-intensive components of the licensing process.

To address these challenges, Idaho National Laboratory (INL) has pioneered the development of an agentic architecture framework leveraging generative artificial intelligence (AI) for advanced reactor safety analysis. This automated AI-driven framework is designed to dramatically enhance accuracy, efficiency, and scalability in performing safety evaluations, thereby accelerating regulatory review and licensing timelines.
This presentation will showcase the first-of-its-kind agentic architecture developed under this research initiative. The framework employs specialized AI “agents,” each optimized for distinct safety analysis tasks, including:

• Failure Modes and Effects Analysis (FMEA)
• Interpretation of multimodal technical data (e.g., diagrams, schematics)
• Assessment of nuclear safety attributes such as initiating events and safety functions
• Construction of fault trees for components and systems

Each agent integrates task-specific AI models, retrieval-augmented generation (RAG) from curated multimodal datasets, and advanced prompt engineering techniques to ensure precision and reliability. The demonstration will detail the workflow architecture, human-in-the-loop validation mechanisms, and outline future research directions aimed at transforming safety analysis into a highly automated, intelligent process.