恶意软件
混淆
计算机科学
对抗制
隐病毒学
操作码
计算机安全
逃避(道德)
利用
人工智能
机器学习
操作系统
免疫系统
免疫学
生物
作者
Dharani Devadiga,Gordon Jin,Bisti Potdar,Hankyu Koo,Andrew Han,Anusha Shringi,Angad Singh,Kinjal Chaudhari,Saurav Kumar
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2023-10-11
卷期号:: 53-58
被引量:28
标识
DOI:10.1109/ictc58733.2023.10393706
摘要
The rapid evolution of cybersecurity threats poses formidable challenges for effective malware detection. Traditional methods often struggle to keep pace with the continuously changing landscape of new malware variants. To address this issue, researchers have turned to machine-learning techniques to optimize malware detection. Yet, these approaches have overlooked the fact that evasive malware is meticulously crafted to elude detection through tactics such as exploiting software vulnerabilities, utilizing encryption, and employing obfuscation techniques. Malware authors have a strong incentive to attack malware detection systems, yet the features and methods that they exploit are limited. We propose a model named GLEAM- GAN and LLM for Evasive Adversarial Malware. This model infuses hex code and opcode features with LLM (Large Language Model) embeddings and GANs (Generative Adversarial Networks) to generate synthetic samples that closely resemble evasive mal-ware to bypass black-box machine learning detectors. Through extensive evaluation, our model achieved an average evasion rate increase of 22.6%, demonstrating its ability to effectively attack detection systems. By expanding the space for adversarial malware generation, we give modern detection systems the capability to counter the nuanced tactics of evasive malware, thus enhancing proficiency in preempting and neutralizing potential threats with heightened precision.
科研通智能强力驱动
Strongly Powered by AbleSci AI