Federated Learning (FL) has emerged as a promising paradigm for privacy-preserving text summarization, allowing decentralized model training on sensitive data without exposing user information. However, conventional FL methods often overlook the necessity of personalization, which is crucial for adapting to diverse user preferences and heterogeneous data distributions. This limitation frequently leads to suboptimal summarization outcomes that do not align well with individual needs. To address this challenge, we introduce a novel FL framework that leverages transformer-based architectures for localized model training while securely aggregating parameter updates at a central server. A core innovation of our approach is a personalized aggregation strategy that explicitly considers client-specific data variations, ensuring improved adaptability of the generated summaries to unique user contexts. Extensive evaluations on widely used benchmark datasets demonstrate that our method not only enhances summarization accuracy but also effectively personalizes outputs. Furthermore, we perform a comparative study of state-of-the-art FL aggregation techniques, highlighting the scalability, flexibility, and practical viability of our approach for real-world text summarization applications. This research contributes to the advancement of FL by addressing both privacy concerns and the critical need for personalized summarization.