Intrusion detection within distributed environments faces significant challenges, primarily stemming from concerns regarding data privacy and the variability of attack patterns. This study presents a novel federated contrastive learning framework aimed at addressing these challenges by combining the strengths of federated learning and contrastive learning methodologies. The proposed framework utilizes a dynamic sparse parameter update strategy to minimize communication overhead while preserving model efficacy. By implementing contrastive learning at the local client level, the framework enhances feature representations, thus improving the model’s ability to distinguish between normal and malicious activities. Experimental findings indicate that the proposed approach substantially outperforms traditional federated learning-based intrusion detection systems in terms of detection accuracy and communication efficiency, thus contributing to the advancement of scalable and privacy-preserving solutions in the field of cybersecurity.