Navigating Cyber Threats

The high-speed development of Fifth Generation technologies announces a new era for the internet of things (IoT), distinguished by high-rate connectivity, speed, and low latency. However, this advancement also opens doors to major security challenges and expands the attack surface. Existing general IoT surveys do not systematically analyze Fifth Generation-enabled IoT concerns, which creates a clear need for a focused synthesis of machine learning and deep learning (DL) defenses tailored to Fifth Generation-IoT constraints and threat models. To address this gap, this survey conducts a preferred reporting items for systematic reviews and meta-analyses-guided analysis of recent studies published in the past 5 years, extracting methodologies, results, datasets, metrics, tools, and reported limitations to answer explicit research questions about which approaches work, under which conditions, and with what deployment implications for Fifth Generation IoT threat detection and mitigation. The results show that DL families and hybrid deep models dominate intrusion, anomaly, and malicious traffic detection, while research overemphasizes denial-of-service attacks relative to Replay, Ransomware, Sybil, Man-in-the-Middle, and Phishing attacks. The recommendations, which come from comparative evidence across datasets, attack categories, and model performance limitations, emphasize the need for more diverse and realistic Fifth Generation IoT datasets as well as for understudied learning paradigms, such as continual learning, federated learning, meta-learning, and self-supervised learning. These insights highlight clear research directions toward adaptive, privacy-preserving, and generalizable intrusion detection in Fifth Generation-IoT systems.