DeepGuardian is an advanced Network Intrusion Detection System designed to safeguard your network by identifying and mitigating security threats in real time. This approach focusses on the security of cloud-native applications by performing anomaly classification over all the inbound and outbound network traffic. The architecture considers the major characteristics of a Federated Learning approach, including the ability to train AI models on decentralized data, thus avoiding potential data breaches that could compromise data security and privacy. By combining a FL-based approach with Convolutional Autoencoders (CAE), DeepGuardian can classify network traffic as normal or malicious without revealing any sensitive information from each application. To increase the existent data privacy level in FL, as well as to increase the trust level between the different federated parties, this framework incorporates advanced privacy-enhancing technologies (PETs) like Zero knowledge Proof (ZKP) and Differential Privacy (DP). The first one is a cryptographic method that enables the verification of data without revealing its content, ensuring that any information remains completely secure and private. This method allows the Aggregator and the Agent to share information without revealing its content, thus increasing data privacy and security throughout training federated rounds. The second one is a technique designed to add controlled noise to the data, ensuring that the output of any analysis does not compromise the privacy of individual data entries. Combine DP with FL-based approach ensure that, after local training and before sharing the model updates between Agent and Aggregator, the data won’t contain any sensitive information about this Agent.