"This high-fidelity dataset comprises over 100 million real-world Controller Area Network (CAN) messages collected directly from a physical Tesla Model 3, providing a comprehensive and representative benchmark for modern Electric Vehicle (EV) network security. While conventional datasets primarily emphasize obvious injection attacks, this collection systematically addresses the critical research gap in stealthy, semantic-level threats. It encompasses a diverse range of attack scenarios, specifically focusing on ID tampering, data field modification, and complex hybrid manipulations. A defining feature of this dataset is the strict preservation of traffic periodicity and frequency, ensuring that the simulated attacks are indistinguishable from legitimate communications via traditional statistical or frequency-based detection methods.Consequently, this dataset provides an exceptionally challenging environment for evaluating the robustness and generalization of next-generation Intrusion Detection Systems (IDS). By offering high-granularity data and sophisticated attack labels, it facilitates the development of advanced deep learning models capable of identifying covert anomalies within safety-critical automotive Cyber-Physical Systems (CPS), ultimately setting a new standard for high-fidelity security validation in the automotive domain."

Figure S1. Determination of the specificity of the MBVs detection by detection of non-MBVs. A panel of non-MBVs were detected by this MBVs detection strategy to determine the specificity. There were no positive results (i.e. detection of these non-MBVs). YFV and EEEV were used as positive references. (XML 7 kb)

Figure S1. Determination of the specificity of the MBVs detection by detection of non-MBVs. A panel of non-MBVs were detected by this MBVs detection strategy to determine the specificity. There were no positive results (i.e. detection of these non-MBVs). YFV and EEEV were used as positive references. (XML 7 kb)