Autonomous vehicles rely on advanced technologies that require various system components to communicate seamlessly with each other. One such technology is the Controller Area Network (CAN). Although widely used in modern vehicles, CAN has various shortcomings and lacks security features. Recent studies have predominantly explored the application of machine learning methods for detecting anomalies. However, this requires a large dataset for proper evaluation and model training. The CrySyS dataset has captured CAN traffic logs in benign and attack states. The dataset focuses on recording traces with varying lengths. It also meets almost all requirements of a detailed dataset required for machine learning algorithms for anomaly detection. Our primary goal with this dataset is to develop algorithms and techniques to detect fabrication and masquerade attacks. This paper presents a pilot study focused on detecting such attacks using machine learning. The study utilized the CrySyS dataset, specifically selecting a trace file that, while unlabeled, contained CAN message signals associated with a known attack type. Through a careful and detailed process, each CAN message was labeled as “Attack” or “Normal” based on observed patterns. The resulting labeled dataset was used to train machine learning models for attack detection. The findings demonstrate the effectiveness of this approach in identifying security threats within autonomous vehicle systems. This study offers valuable insights into the critical role of accurate data labeling in developing robust intrusion detection systems (IDSs) for the future of autonomous transportation.