The rapid growth of digital financial transactions and the widespread adoption of financial messaging systems have created new opportunities for sophisticated and evolving fraud patterns. Traditional fraud detection approaches often face significant limitations due to centralized data dependency, privacy concerns, and inadequate adaptability to dynamic fraud behaviors. This study proposes a federated distributed learning architecture based on stochastic artificial intelligence for identifying evolving fraud patterns in financial messaging systems. In the proposed framework, financial data remain locally stored at participating institutions, while only model parameters are exchanged across nodes, thereby preserving data privacy and regulatory compliance. Furthermore, the integration of stochastic artificial intelligence techniques enables the model to dynamically adapt to behavioral changes and emerging fraud schemes. This research is applied in purpose and developmental in methodology, relying on conceptual modeling and intelligent system design. The expected results suggest that the combination of federated learning and stochastic artificial intelligence can significantly enhance fraud detection accuracy and responsiveness while reducing operational, privacy, and compliance risks in financial institutions. The proposed architecture provides a novel framework for developing intelligent fraud detection systems within modern financial and banking infrastructures.