Artificial intelligence (AI) technology, a key driver leading the latest technological advancements, is profoundly transforming transportation systems. While AI significantly enhances system efficiency and service capabilities, it also introduces unprecedented security risks. These risks stem from inherent vulnerabilities in AI models and data, as well as the complexity of their integration into open and dynamic traffic environments. Specific manifestations include: model inexplicability, insufficient generalization ability, data bias, amplification of cyber attacks, and physical control failure. This paper analyzes the security challenges of AI applications in intelligent transportation,and proposes a hierarchical protection framework comprising “Basic Universal Security Capabilities” and “Scenario-Enhanced Security Capabilities”. Corresponding solutions are presented in conjunction with typical scenarios. This research aims to provide theoretical support and practical pathways for enhancing the overall robustness of intelligent transportation systems, thereby supporting the development of next-generation intelligent transportation.