Federated Learning and Confidential Computing for Privacy-Preserving Cross- Organizational Analytics

Abstract

Cross-organizational analytics requires a privacy architecture that can support shared model learning without moving sensitive data into a central repository. This article presents a cloud-native framework that combines federated learning, confidential computing, secure aggregation, and automated compliance reporting for privacy-preserving collaboration among regulated organizations. The architecture keeps raw data inside participant environments, protects model-update aggregation through trusted execution and encrypted transfer, and converts each workflow event into verifiable compliance evidence. Results show that federated model accuracy, privacy leakage reduction, and compliance evidence completeness improve across training rounds, while secure aggregation reliability remains high across participating organizations despite confidential computing overhead and reporting latency. The study shows that federated analytics becomes more practical when privacy protection and audit reporting are designed as one continuous workflow.