In order to address the privacy leakage and audit untrace ability issues of corporate financial data in multi-organization collaboration, a privacy-preserving modeling and auditing algorithm based on the federated learning framework was constructed. This method combines differential privacy with verifiable computing technology to achieve privacy shielding and audit visualization control of the model update process. Based on the theoretical model, simulation experiments were designed to evaluate its comprehensive performance in data security, model accuracy and auditing capabilities. The experimental results show that the proposed algorithm significantly improves the robustness and audit transparency of the system while ensuring data localization, and has higher security and compliance advantages than the traditional centralized model. The research provides a feasible technical path for corporate financial information sharing, and also lays a theoretical foundation for the trusted construction of intelligent financial systems.