Federated Learning in Cloud-Native Architectures: A Secure Approach to Decentralized AI

Purpose: The paper aims to analyze the technical and security challenges of deploying FL at scale and explores how modern cloud-native technologies such as container orchestration, hybrid cloud infrastructure, and privacy-preserving techniques can be leveraged to mitigate these challenges. The study also seeks to provide a comprehensive understanding of how FL is being applied in critical domains such as healthcare, IoT, and cybersecurity, while identifying future trends that could shape the evolution of decentralized AI systems. Methodology: This research adopts a qualitative and architectural analysis approach to evaluate the intersection of Federated Learning and cloud-native computing. A systematic review of the current state-of-the-art technologies supporting FL, including Docker containers, Kubernetes orchestration, and hybrid cloud environments. A threat modeling analysis focusing on prevalent security risks such as data poisoning, model inversion, and Byzantine node attacks. An evaluation of security frameworks and privacy-enhancing technologies (e.g., differential privacy, secure multi-party computation, and homomorphic encryption) used to protect FL systems. Findings: The study finds that cloud-native architectures provide a robust and flexible foundation for scaling Federated Learning systems. Kubernetes-based orchestration and containerization significantly enhance the deployment and scalability of FL models across heterogeneous environments. Unique Contribution to Theory, Practice and Policy: While FL minimizes raw data exchange, it introduces unique attack vectors; effective mitigation requires multi-layered security, including encryption protocols and node validation mechanisms. Techniques such as differential privacy and homomorphic encryption provide meaningful protections but must be carefully balanced against performance overhead.