In the rapidly evolving landscape of vehicular communications, the widespread use of the Controller Area Network (CAN) in modern vehicles has revealed significant security vulnerabilities. However, existing Intrusion Detection Systems (IDS) struggle to adapt to varied attack scenarios and precisely detect low-volume attacks. In this paper, we introduce a novel IDS that employs meta-learning via the Meta-SGD algorithm, enhancing adaptability across a diverse spectrum of cyber threats, called Meta-IDS. Specifically, our methodology includes a bi-level optimization technique where the inner level focuses on optimizing detection accuracy for specific attack scenarios, and the outer level adjusts meta-parameters to ensure generalizability across different scenarios. For modeling low-volume attacks, we devise the Attack Prominence Score (APS), identifying subtle attack patterns with a threshold of APS \(\textgreater\) 7, allowing for precise differentiation of these attacks. The extensive experiment results show that the proposed method facilitates efficient tuning and rapid adaptation for different modeling paradigms in few-shot scenarios. The detection performance is exceptional, with F1-scores reaching 100% across most attack scenarios, including low-volume attacks. Also, the real-time vehicle-level evaluations demonstrate its adaptability for the vehicular networks.