Abstract— Smart transportation systems empowered by Internet of Things (IoT) devices and artificial intelligence (AI) are increasingly vulnerable to adversarial attacks that can compromise safety, privacy, and operational efficiency. This paper presents TrustGuard, a novel multi-layered defense framework specifically designed to ensure adversarial robustness in IoT-based transportation networks. The primary objective is to detect and mitigate adversarial perturbations while maintaining real-time performance on resource-constrained IoT devices. We propose a hierarchical defensive architecture that combines lightweight adversarial detection at the edge with more complex defense mechanisms in the fog layer. Our methodology introduces an adaptive perturbation filtering technique that achieves 92.6% detection accuracy for common adversarial attacks with minimal computational overhead (23ms average processing time on edge devices). Extensive experiments conducted on three public transportation datasets (GTSRB, CityFlow, and BDD100K) demonstrate that TrustGuard outperforms state-of-the-art defense mechanisms by an average of 18.3% in detection accuracy and 37.4% in processing efficiency. Statistical analysis using Mann-Whitney U tests (p < 0.01) confirms the significance of our results. The framework’s ability to maintain transportation system performance under various attack scenarios while operating within the resource constraints of IoT environments represents a significant advancement in developing trustworthy AI for smart transportation infrastructure.