The Sixth Generation Heterogeneous Network (6G HetNet) is a global interconnected system that serves a myriad variety of applications and services across multiple domains such as satellite, air, ground, and underwater networks. It provides a platform for the development of novel Internet of Things (IoT) applications and services, particularly for the Internet of Vehicles (IoV), which encompasses all devices involved in intra-vehicle and inter-vehicle communications. However, this evolution towards a unified and huge cellular infrastructure creates new security challenges that require an intelligent attack detection framework to safeguard the network against cyber-security threats. This paper proposes a hierarchical attack detection framework for 6G-enabled IoV. This framework relies on the processing capacities of edge nodes to satisfy the main 6G Key Performance Indicators (KPIs), such as trustworthiness, latency, connectivity, data rate and energy consumption. Federated Learning (FL) and non-cooperative gaming are used to train attack models and improve the detection process over time. The cooperative detection process based on FL is executed by security entities, IoV devices, edge servers and Security Information and Event Management (SIEM) to improve the detection accuracy over time. To harden the security of the proposed attack detection framework, a robust Stackelberg security game is developed to identify malicious IoV devices and edge servers, and select suitable IoV devices and edge servers to participate in the training and attack detection processes. The identification and selection process mainly relies on computing a reputation score based on the activities of these IoV devices and edge servers. As compared to current security monitoring and detection solutions, our framework balances detection accuracy and reduced network overhead, specifically as the system scales up, i.e., when the malicious traffic is high. In addition, it mitigates threats from both external and internal adversaries.