Abstract

Despite the benefits of Internet of Things (IoT) devices, the insecurity of IoT and their deployment nature have turned them into attractive targets for adversaries, which contributed to the rise of IoT-tailored malware as a major threat to the Internet ecosystem. In this thesis, we address the threats associated with the emerging IoT malware, which utilize exploited devices to perform large-scale cyber attacks (e.g., DDoS). To mitigate such threat, there is a need to possess an Internet perspective of the deployed IoT devices while building a better understanding about the behavioral characteristic of malware-infected devices, which is challenging due to the lack of empirical data and knowledge about the deployed IoT devices and their behavioral characteristics.

To address these challenges, in this thesis, we leverage passive Internet measurements and IoT device information to detect exploited IoT devices and investigate their generated traffic at the network telescope (darknet). We aim at proposing data-driven approaches for effective and near real-time IoT threat detection and characterization. Additionally, we leverage a specialized IoT Honeypot to analyze a large corpus of real IoT malware binary executable. We aim at building a better understanding about the current state of IoT malware while addressing the problems of IoT malware classification and family attribution. To this end, we perform the following to achieve our objectives:

First, we address the lack of empirical data and knowledge about IoT devices and their activities. To this end, we leverage an online IoT search engine (e.g., Shodan.io) to obtain publicly available device information in the realms of consumer and cyber-physical system (CPS), while utilizing passive network measurements collected at a large-scale network telescope (CAIDA), to infer compromised devices and their unsolicited activities. Indeed, we were among the first to report experimental results on detecting compromised IoT devices and their behavioral characteristics in the wild, while demonstrating their active involvement in large-scale malware-generated malicious activities such as Internet scanning. Additionally, we leverage the IoT-generated backscatter traffic towards the network telescope to shed light on IoT devices that were victims of intensive Denial of Service (DoS) attacks.

Second, given the highly orchestrated nature of IoT-driven cyber-attacks, we focus on the analysis of IoT-generated scanning activities to detect and characterize scanning campaigns generated by IoT botnets. To this end, we aggregate IoT-generated traffic and performing association rules mining to infer campaigns through common scanning objectives represented by targeted destination ports. Further, we leverage behavioural characteristics and aggregated flow features to correlate IoT devices using DBSCAN clustering algorithm. Indeed, our findings shed light on compromised IoT devices, which tend to operate within well coordinated IoT botnets.

Third, considering the huge number of IoT devices and the magnitude of their malicious scanning traffic, we focus on addressing the operational challenges to automate large-scale campaign detection and analysis while generating threat intelligence in a timely manner.
To this end, we leverage big data analytic frameworks such as Apache Spark to develop a scalable system for automated detection of infected IoT devices and characterization of their scanning activities using our proposed approach. Our evaluation results with over 4TB of IoT traffic demonstrated the effectiveness of the system to infer scanning campaigns generated by IoT botnets. Moreover, we demonstrate the feasibility of the implemented system/framework as a platform for implementing further supporting applications, which leverage passive Internet measurement for characterizing IoT traffic and generating IoT-related threat intelligence.

Fourth, we take first steps towards mitigating threats associated with the rise of IoT malware by creating a better understanding about the characteristics and inter-relations of IoT malware. To this end, we analyze about 70,000 IoT malware binaries obtained by a specialized IoT honeypot in the past two years. We investigate the distribution of IoT malware across known families, while exploring their detection timeline and persistent. Moreover, while we shed light on the effectiveness of IoT honeypots in detecting new/unknown malware samples, we utilize static and dynamic malware analysis techniques to uncover adversarial infrastructure and investigate functional similarities. Indeed, our findings enable unknown malware labeling/attribution while identifying new IoT malware variants. Additionally, we collect malware-generated scanning traffic (whenever available) to explore behavioral characteristics and associated threats/vulnerabilities.

We conclude this thesis by discussing research gaps that pave the way for future work.