This massive scale of deployment of IoT-enabled devices such as traffic sensors, surveillance cameras, and utility meters in smart cities has resulted in fast growth. Even though this has improved the efficiency and better management of urban systems, the connected devices were more vulnerable to cybersecurity threats such as unauthorized access, data breaches, and DDoS attacks. This research advanced to propose the utilization of ML and AI algorithm functionalities for detecting and mitigating similar kinds of threats within an IoT environment. Four models artificial neural network (ANN), support vector machine (SVM), random forest (RF), and recurrent neural network (RNN) are developed and tested toward examining ability to identify and respond. The results were achieved by the RNN model in the maximum accuracy of 96.4, followed by the ANN model with an accuracy of 93.2, then SVM at 90.5, and 92.3% by RF. In terms of performance, it was much better by the RNN with the latency of 50 ms and the throughput of 200 operations per second. However, ANN demonstrated the latency of 45 ms, and the throughput of 220 operations per second. The SVM and RF models showed higher latencies of 65 ms and 70 ms with lower throughputs of 180 and 160 operations per second, respectively. These results indicate the promise of deep learning models in the event of their deployment in real-time cybersecurity domains of IoT devices. Such successful deployment of models can make the security effectiveness of IoT networks in smart cities take a leap toward safe and efficient urban infrastructure.