김명현, 조해현, 서창배, 임유빈, 강성현, 조병모,

Vol. 35, No. 2, pp. 265-275, 4월. 2025
10.13089/JKIISC.2025.35.2.265, Full Text:
Keywords: Malware, deeplearning
Abstract

Today, the use of artificial intelligence is endless and is used in various fields. Among them, artificial intelligence models are used in the security field to detect malicious files and prevent damage caused by the malicious files. In this study, a framework for performing hostile attacks and analyzing vulnerabilities by dividing into gradient-based and gradient-free situations for specific malicious file detection models and defense techniques for malicious file detection models are proposed and applied. As a result of the experiment, it showed an attack success rate of up to 98% against attack and a maximum attack success rate of 2% after applying the defense technique. It was confirmed that a vulnerability analysis report was created in this process. Using the framework calculated in this paper, we intend to more effectively utilize artificial intelligence in the security field and secure the stability of the existing malicious file detection model and the stability of unspecified models.

Statistics
Cite this article