An Infection-Identifying and Self-Evolving System for IoT Early Defense from Multi-Step Attacks

Abstract

AbstractInternet-of-Things (IoT) cyber threats such as jackware [14] and cryptomining [33] show that insecure IoT devices can be exploited by attackers with different goals. As many such attacks are multi-steps, early detection is critical. Early detection enables early attack containment and response, and prevention of malware propagation. However, it is challenging to detect early-phase attacks with both high precision and high recall as attackers typically attempt to evade the detection systems with stealthy or zero-day attacks. To enhance the security of IoT devices, we propose IoTEDef, a deep learning-based system able to identify the infection events and evolve with the identified infections. IoTEDef understands multi-step attacks based on cyber kill chains and maintains detectors for each step. When it detects anomalies related to a later stage of the kill chain, IoTEDef backtracks the log of events and analyzes these events to identify infection events. Then, IoTEDef updates its infection detector with the identified events. IoTEDef can be used for threat hunting as well as the generation of indicators of compromise and attacks. To show its feasibility, we implement a prototype of the system and evaluate it against the Mirai botnet campaign [2] and the multi-step attack that exploits the Log4j vulnerability [36] to infect the IoT devices. Our results show that the F1-score of our evolved infection detector in IoTEDef, instantiated with long short-term memory (LSTM) and the attention mechanism, increases from 0.31 to 0.87 . We also show that existing attention-based NIDSes can benefit from our approach.KeywordsInternet of thingsMulti-step attacksInfection identificationThreat huntingAttention mechanism