JaeHwan Kwon, JuHo Bai, Gwang-Yong Gim, Journal of Internet Computing and Services, Vol. 26, No. 3, pp. 135-144, Jun. 2025
10.7472/jksii.2025.26.3.135, Full Text:  HTML
Keywords: Retrieval Augmented Generation(RAG), Large Language Model(LLM) Agent, Security Guide, Security Intelligence

Abstract

In this study, we propose the SIREN (Security Intelligence & Regulation Enforcement Network) model, which integrates Retrieval Augmented Generation (RAG) and large language models (LLMs) to provide an intelligent security guide. By adopting an LLM agent–based control mechanism, SIREN reduces security threats such as prompt injection attacks and sensitive data leaks that may arise when deploying LLMs in internal environments. We built a prototype model and conducted experiments with various scenarios and user roles simulating internal security breaches. The experimental results show that SIREN improves recall compared to traditional rule-based security systems, achieving around a 40% increase in F1 scores, and no sensitive data leakage occurred in the LLM-based SIREN pipeline. This indicates that SIREN has practical potential for applying artificial intelligence in secure environments, including the defense sector.

Statistics
Cite this article