Vehicular Ad Hoc Networks (VANETs) require robust security mechanisms to detect and mitigate node misbehavior and cyber-attacks that can jeopardize traffic safety. Traditional Intrusion Detection Systems (IDS) for VANETs face challenges in balancing detection accuracy with real-time constraints and resource limits on vehicles. In this paper, we propose a trust-gated hybrid IDS architecture that combines lightweight on-board unit (OBU) screening with a more powerful roadside unit (RSU) supervised fusion. Each vehicle (OBU) performs quick plausibility and consistency checks on received messages, flagging anomalies with negligible delay. Suspicious evidence is then processed by an RSU-based IDS that integrates multiple machine learning detectors specialized for different attack families (e.g., position tampering, replay, denial-of-service, Sybil, and protocol abuses). A Bayesian trust management module continuously updates a trust score for each vehicle based on its behavior, and dynamically adjusts detection thresholds per sender, a process we term “trust gating.” This adaptive thresholding reduces false alarms by accounting for historical honesty or misbehavior of nodes. We implement the proposed system using efficient machine learning models (e.g., LightGBM, calibrated logistic regression stacking, and an LSTM for temporal attacks) to ensure real-time performance. Evaluation on the standard VeReMi misbehavior dataset and realistic simulation scenarios (urban VANET setups in Baghdad via the F2MD framework) demonstrates that our hybrid approach achieves high detection rates (with macro F1-scores = 0.979, AUC(avg) = 0.9822 and Accuracy = 0.9971) and low false positive rates, outperforming baseline IDS methods including one-class SVM, Isolation Forest, Local Outlier Factor, and single-stage classifiers (Random Forest, XGBoost, SVM, and deep CNN/GRU models). The trust gating mechanism notably improves precision-recall tradeoff by tailoring sensitivity to each vehicle’s trust level. Our results indicate that the proposed IDS can significantly enhance security in VANETs while meeting real-time and deployment constraints. Code and dataset are publicly available to support full reproducibility.