Although the potential of Artificial Intelligence (AI) in streamlining the integration of Terrestrial Networks (TNs) and Non-Terrestrial Networks (NTNs) has captivated significant research interest, its reliability in delivering and maintaining the required Quality of Service (QoS) in such highly heterogeneous, dynamic, and stochastic environments is often overlooked. In this section, we present a reliable AI framework for unified optimization and management of resources on the ground and in various orbits. The framework aims to address the challenges that hinder the reliability of AI solutions, which are deeply intertwined with the inherent complexities of integrated TNs/NTNs. First, we identify the architectural characteristics and challenges of integrated TNs/NTNs that directly impact AI performance, and define the key aspects that influence AI reliability in these environments by reviewing state-of-the-art works. We then introduce the proposed reliable AI framework, detailing the key processes and steps that should be adopted across all phases of the AI lifecycle to ensure and sustain reliability. Finally, we present a case study on computation task offloading in integrated TNs/NTNs, illustrating how individual mechanisms from the proposed reliable AI framework can be applied to enable reliable decision-making. Direct extension to other decision-making and resource allocation problems can be readily pursued, paving the way for future research.