Abstract
The rapid growth of the Internet of Things (IoT) devices is paralleled by them being on the front-line of malicious attacks caused by malicious software. Machine learning (ML) algorithms, alongside the traditional signature-based methods, are typically used to detect malicious activities and behaviors. However, they are susceptible to malware evolution and sophistication, making them limited to the patterns that they have been trained upon. In this work, we systematically examine the state-of-the-art malware detection approaches using various representations, under a range of adversarial settings. Our preliminary analyses highlight the instability of the learning algorithms in learning patterns that distinguish the benign from the malicious. Our mutations with functionality-preserving operations, e.g., software stripping and binary padding, significantly deteriorate the accuracy of malware detectors.