김승현, 이상진,

Vol. 35, No. 2, pp. 323-334, 4월. 2025
10.13089/JKIISC.2025.35.2.323, Full Text:
Keywords: LLM, SQL Injection, Mutation, Detection
Abstract

The recent advancement of large language models (LLM) is also presenting new possibilities in the security field based on code-level contextual understanding and generation capabilities. In this study, we use these LLMs to test the potential as a mutation engine and detection engine to efficiently detect SQL injection vulnerabilities in web application environments. Specifically, LLM is used as a mutation engine to automatically generate sophisticated attack payloads that fit the actual SQL context beyond simple string variation, and LLM is used as a detection engine to verify that changes in SQL context are recognized and vulnerabilities are immediately identified. As a result of the experiment, it was confirmed that this approach can be used as a more effective security technology by understanding the SQL context and capturing various injection patterns.

Statistics
Cite this article