Urban traffic congestion can be considered as a significant problem, and it contributes to long travel periods, fuel usage, and environmental influence. This paper introduces an Intelligent Traffic Control System (ITCS) that consists of Vehicular Ad Hoc Networks (VANETs) and Reinforcement Learning (RL) to optimise the control of traffic signals. The system facilitates real-time two-way communication between vehicles and roadside units, which means that an RL agent can control signal phases adaptively according to the traffic metrics like the average delay, the queue length, and traffic throughput. The Kaggle VANET Malicious Node Dataset was used to simulate malicious or unreliable nodes and test the robustness of the systems. The RL agent has been trained on the SUMO simulator trained on TraCI through various episodes and learns to take actions that increase traffic movement with a minimum amount of congestion. The results of training are progressive, as cumulative rewards grow, and average delays and queue length reduce with epochs. Performance evaluation of the ITCS under peak-hour, off-peak, incident, and malicious node scenarios demonstrated substantial gains over conventional fixed-time controllers, with average delays reduced by 48–55%, queue lengths by 49–57%, and throughput increased by 28–35%. These results indicate the success of the blend of reinforcement learning with VANET-supported traffic control, which is an adaptive, data-driven, and robust solution to an urban intersection. Not only the RL-based ITCS enhances traffic flow and congestion, but is also resistant to communication anomalies, which indicates its scalability to be deployed in the current smart city traffic management.