The rapid proliferation of Internet of Things (IoT) devices, advanced edge computing, and high-speed telecommunications, which have catalysed a profound transformation in urban infrastructure and the evolution of the modern smart city. Central to this transformation is the deployment of Intelligent Transportation Systems (ITS) and Smart City Traffic Management Systems (SCTMS). These sophisticated architectures leverage machine learning algorithms, pervasive sensor networks, and 5G connectivity to dynamically optimize vehicular flow, alleviate urban congestion, and significantly enhance public safety metrics. However, the deep integration of physical civic infrastructure with decentralized digital networks introduces an unprecedented and expanding cybersecurity attack surface. As municipalities transition to deeply interconnected, autonomous networks, the risk from malicious actors—ranging from cybercriminals to advanced persistent threats—increases. Potential threat vectors include Distributed Denial of Service (DDoS) attacks, data theft, and the hijacking of edge sensors. These developments warrant further discussion as we consider the future of our digital and physical security This comprehensive research report regarding the security of smart traffic systems and the emerging threats posed by Adversarial Machine Learning (AML). The report provides an exhaustive analysis of architectural frameworks and the vulnerabilities introduced by advanced techniques, such as data poisoning and evasion attacks. Furthermore, the research evaluates ML-driven defensive paradigms—including high-dimensional anomaly detection and federated learning—and features an extensive case study on the Saudi Data and Artificial Intelligence Authority (SDAIA) and the Sawaher platform deployed in Riyadh and during the Hajj pilgrimage. Ultimately, the report proposes a multi-layered architectural solution integrating Zero Trust Architecture (ZTA) with adversarial robust Deep Reinforcement Learning (DRL) to ensure urban infrastructure remains resilient against next-generation algorithmic subversion.