Quantum-Aware Privacy Challenges in Differentially Private Federated Learning

Federated learning (FL) enables multiple parties to collaboratively train models without sharing their raw data, making it suitable for environments where data confidentiality is important. Many FL systems rely on differential privacy (DP) to protect sensitive information, mainly focusing on balancing privacy protection and model performance. However, with the rapid growth of computational capabilities and increasingly sophisticated inference attacks, the long-term effectiveness of these approaches is becoming uncertain. This work reviews the development of privacy-related issues in DP-based FL from 2010 to 2026, focusing on three key areas: privacy protection mechanisms, information leakage caused by attacks, and the role of adversaries. A review of existing studies shows that these aspects have largely been addressed independently, while a comprehensive framework that jointly evaluates privacy, information leakage, and model performance is still lacking.