Federated learning is a distributed approach to machine learning and statistics where models are trained across multiple decentralised devices or servers holding local data samples. It's often perceived as a privacy- enhancing technique because it avoids centralising sensitive data. However, federated learning itself does not inherently guarantee privacy. While it reduces the risk of data exposure by keeping data on local devices, the process of aggregating model updates can potentially reveal sensitive information. Therefore, federated learning must be combined with explicit privacy-preserving mechanisms, such as differential privacy, to ensure that individual data contributions cannot be inferred from the shared model updates.