This chapter explores how federated learning (FL) systems can be designed to meet key requirements for trustworthy artificial intelligence. It draws on international policy frameworks, including those from the EU, USA, OECD, and others. The chapter begins by addressing human agency and oversight, emphasizing model simplicity, continuous validation, and ethical limitations on data use. It then analyzes the technical robustness of FL systems, including their sensitivity to data perturbations, estimation errors, and infrastructure faults. Robustness is evaluated for both the learned models and the algorithms that compute them. The chapter also highlights strategies to improve network resilience and prevent single points of failure. Privacy and data governance are discussed in light of legal constraints, with a focus on minimizing unnecessary data use and ensuring organizational accountability. The chapter closes by discussing transparency, explainability, fairness, and environmental sustainability. It proposes ways to enforce explainability through user-specific penalties in the optimization objective. Fairness is addressed through data augmentation, while societal and environmental impacts are acknowledged through system design choices. Altogether, the chapter builds a bridge between mathematical methods like GTVMin and broader concerns about AI ethics and accountability.