Federated learning (FL) enables collaborative model training across multiple clients without sharing raw data, offering a promising solution for privacy-sensitive applications. However, as FL becomes more decentralised, balancing data privacy with resilience against adversarial attacks remains a fundamental challenge. This thesis investigates the interplay between privacy-preserving mechanisms such as Differential Privacy, Secure Multi-Party Computation (SMPC), and Subspace Perturbation, and the robustness of adversarial detection in fully decentralised FL networks. By extending information-theoretic bounds and conducting comprehensive experiments under a variety of attack scenarios, we show that stronger privacy guarantees often come at the cost of reduced detection capability. Notably, mechanisms that increase noise or mask updates to protect data privacy tend to obscure the test statistics that detectors rely on, resulting in higher false alarm rates and missed detections. Our results highlight that while privacy and robustness cannot be maximised simultaneously, careful tuning of system parameters and defence strategies can help achieve a practical balance. This work provides theoretical insights and empirical evidence to inform the deployment of privacy-preserving and robust federated learning systems.