Real-time communication between autonomous vehicles, infrastructure, and their environment has facilitated the Internet of Vehicles (IoVs). Although this connectivity provides vehicular networks with significant benefits, it also introduces severe security threats, such as message injection attacks, particularly due to the heavy reliance on the Controller Area Network (CAN) protocol, which is inherently vulnerable. Electronic Control Units (ECUs) become primary targets for these attacks, leading to unsafe vehicle behaviors. To address these challenges, an Intrusion Detection System (IDS) based on deep learning architectures, Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU) is proposed for detecting and classifying cyberattacks in vehicular networks. Data preprocessing techniques such as NearMiss, Random Over-Sampling (ROS), and Tomek Links are applied to handle class imbalance in the car hacking dataset. A benchmark Car-Hacking dataset, evaluation metrics including accuracy, precision, recall, and F1 score, are used to assess the performance of the models. The experimental results demonstrate that the GRU model achieves the highest accuracy of 99%, followed by LSTM with 98%, and RNN with 94%. These findings indicate that GRU outperforms the other models and, in certain configurations, detects 100% of the attacks. The proposed IDS exhibits considerable superiority over traditional deep learning approaches, presenting a promising and intelligent solution for enhancing the cybersecurity of modern IoV systems.