Full Program
Summary:
In this work, we present a novel approach for generating adversarial attacks on malware classification systems that rely on image-based representations of binary executables. Our method selectively applies obfuscation techniques to modify specific bytes in the binary, which correspond to adversarially perturbed pixels in the representation of malware as an image. By leveraging syntactic obfuscation strategies, we are able to transform the malware binary without compromising its functionality. Our results demonstrate that our approach effectively fools the CNN-based detection techniques, leading to misclassification. Additionally, we address the challenges associated with selective obfuscation, particularly when modified bytes map to non-instructional regions or structural elements of the binary. Overall, this research opens new avenues for understanding and defending against adversarial attacks in malware detection systems.Author(s):
Claudia Greco
University of Calabria
Italy
Michele Ianni
University of Calabria
Italy
Antonella Guzzo
University of Calabria
Italy
Giancarlo Fortino
University of Calabria
Italy