Creating Adversarial Malware Examples Through Guided Metamorphic Changes

Malconv is considered the state-of-the-art in mal-ware detection for Windows PE executables. It is a Convolutional Deep Learning model that is capable of consuming executable byte code to detect malware with high accuracy. Attacks have been created that can effectively generate adversarial examples to evade the detection of Malconv. However, these attacks generate adversarial examples by inserting benign bytes to maintain the malware’s malicious functionalities, an approach that is presumed to be easily detectable. In this paper, a novel black-box attack is proposed to create adversarial examples against Malconv, where functionally-equivalent instruction replacements are combined with Particle Swarm Optimization to create adversarial examples. The overall goal is to demonstrate that Malconv is prone to sophisticated attacks that interact with the executable portions of the malware.