The nuclear safeguards regime is facing a critical challenge of information overload, where the volume of data from inspections, state declarations, and open sources threatens to exceed finite analytical resources. Generative Artificial Intelligence (GenAI) and Large Language Models (LLMs) offer a transformative opportunity to address this issue by automating the analysis of vast, unstructured datasets. This paper provides a comprehensive review of the application of GenAI to nuclear safeguards. It examines the institutional frameworks for AI governance being developed by bodies like the IAEA and Euratom and situates these efforts within the broader context of GenAI's impact on Open Source Intelligence (OSINT). The paper explores two primary deployment models for AI, namely general-purpose ‘horizontal’ platforms and specialized ‘vertical’ applications, and discusses the strategic considerations for their implementation. Finally, it presents the results of two demonstrator projects: a Retrieval-Augmented Generation (RAG) system for question-answering over a specialized news database, and an LLM-based workflow for filtering and analyzing large volumes of social media data. The findings highlight the potential of GenAI to serve as an analyst's “co-pilot”, enhancing the efficiency and effectiveness of safeguards information analysis while retaining human review and decision-making.

ANGORRILLA BUSTAMANTE Carlos;  WIGHTMAN Maria-Louisa;  WOLFART Erik;  RENDA Guido; 

2026-09-17

PUBLICATIONS OFFICE OF THE EUROPEAN UNION

JRC146582

1977-5296 (online),   

https://publications.jrc.ec.europa.eu/repository/handle/JRC146582,   

10.3011/ESARDA.IJNSNP.2026.4 (online),