The open radio access networks (open RAN) initiative, e.g., O-RAN, shows sweeping momentum in shaping, revolutionizing, and defining next-generation 6G mobile networks. As the key focus of 6G, network autonomy is much anticipated to autonomously manage, operate, and optimize real-world networks, particularly leveraging Artificial Intelligence / Machine Learning (AI/ML) techniques. Existing domain-agnostic AI/ML techniques fail to safely adapt and generalize under unforeseeable, evolving network dynamics, which has become one of the foremost stumbling blocks for the telecommunication industry to fully embrace AI-assisted network automation toward next-generation networks. The research goal of this CAREER project is to achieve trustworthy network autonomy by optimizing for safety, generalizability, and interpretability. To this end, domain-informed AI/ML innovations will be developed to address generic online network control problems in ever-evolving real-world mobile networks. The education goal is to develop next-generation STEM workforce, cultivate and retain talents locally, and engage and inspire K-12 students. The integrated research and education will contribute to bridging the digital divide in Nebraska, rural America, and beyond. The research program of this CAREER project will derive a novel systematic framework of trustworthy AI-native network autonomy in open RAN. First, new online task-oriented digital network twin (DNT) frameworks will be designed to derive DNTs with all the attributes of fidelity, synchronicity, and tractability. Second, new safe deep reinforcement learning (DRL) frameworks will be designed to achieve verifiable safety for online resource allocation (e.g., dApps/xApps) in real-world networks. Third, new explanation-guided Bayesian optimization frameworks will be designed to achieve interpretable safety for online network configuration (e.g., rApps) under time-evolving dynamics. Moreover, the education program will develop a new camp