Abstract
Financial institutions are currently faced with suffering never experienced before as they strive to guarantee the privacy of data and address the demands of regulation to report and cooperate in machine learning. This paper proposes PrivChain-AI, a novel blockchain-based federated learning system designed to facilitate secure and privacy-preserving financial reporting and access control. The proposed framework will integrate three key components: differential privacy, homomorphic encryption, and smart contract-based governance, enabling cooperative model training across financial institutions while preventing the leakage of sensitive information. PrivChain-AI is a hierarchical design that incorporates permissioned consensus protocols and utilises zero-knowledge proof verification to authenticate transactions. It has been demonstrated that the performance is higher than that of the actual financial data, with an outcome of 94.7% accuracy in fraud recognition at the cost of e-differentiation privacy, where ϵ = 1.0. It is 40% faster in terms of communication overhead and ensures regulatory compliance, as it features immutable audit trails. The analysis of performances reveals that a privacy preservation metric improves by 78%, and access control granularity is improved by 62% compared to the current state-of-the-art approaches. The PrivChain-AI paradigm introduced provides a new analytical model for safe, collaborative finance, meeting the highest standards and ensuring compliance with relevant regulatory jurisdictions.