Smart cities are becoming increasingly popular with the advancement of the Internet of Things (IoT). A vast amount of information is generated during IoT communications, influencing services such as e-governance, intelligent transportation, and smart education. However, most existing security measures focus primarily on data security rather than securing the entire data transmission process. Ensuring the security of IoT devices is critical, as smart cities rely on automated IoT systems for various tasks. A major challenge lies in the identification and authentication of nodes, along with the detection of sophisticated cyberattacks in real time. The rapid growth of IoT continues to raise privacy and security concerns, exacerbated by limited battery life and scalability constraints. To address these challenges, this research proposes a hybrid security framework combining lightweight Elliptic Curve Cryptography (ECC) for efficient node authentication and a CNN-GRU hybrid deep learning model for real-time anomaly detection. The proposed framework is trained and evaluated on the Edge-IIoTset cyber security dataset, covering multiple types of attacks. Experimental results demonstrate that the model achieves high detection accuracy with reduced computational overhead, making it suitable for deployment in resource-constrained IoT environments. This study aims to guide researchers and developers toward more resilient, efficient, and scalable security mechanisms for IoT driven smart infrastructures.