Abstract:With the rapid development of the Internet of Things technology and the emerging need for small and efficient equipment, IoT devices are attracting huge attention. The increasing integration of IoT devices into practical applications has not only heightened their appeal but also raised significant concerns. Despite their ease of deployment and cost-effectiveness, the security measures of these devices have not kept pace with their expanding use. In an era where cyber-attacks are becoming increasingly sophisticated, enhancing the security of IoT devices is imperative to thwart malicious efforts by hackers and to ensure cost savings. In this paper, we present a novel dynamical cross-architecture IoT malware detection model that utilizes IoT software's dynamical behaviors such as system calls to achieve a highly accurate detection rate. A comprehensive evaluation is performed for the proposed model which finally achieved a high score of 99.44\% average accuracy detecting unknown IoT Executable and Linkable Format (ELF). Compared with other methods for IoT malware detection, our method is easy to deploy and able to achieve a high rate of detection, which makes it suitable for defending malicious IoT software and protecting the integrity and security of IoT ecosystems.