Federated learning (FL) has emerged as a transformative paradigm for collaborative model training without the need to centralize sensitive information. By enabling multiple participants to train a shared model locally and only exchange model updates, FL preserves privacy while leveraging the diversity of distributed data. This approach is particularly significant in domains such as healthcare, finance, and industrial Internet of Things, where data confidentiality and compliance with regulatory standards are critical. Despite its promise, FL faces challenges related to security vulnerabilities, communication overhead, and model aggregation fairness across heterogeneous networks. Recent advances in secure aggregation, differential privacy, and blockchain integration have shown potential in mitigating these risks while ensuring trust among participants. This paper examines the role of federated learning as a mechanism for secure data sharing across distributed networks, highlighting its core advantages, limitations, and future directions for achieving scalable and resilient decentralized intelligence.