BLOCKCHAIN-ENABLED INFRASTRUCTURE FOR SECURE AND AUDITABLE AI SYSTEMS

Abstract

Security and auditability are critical requirements for AI systems deployed in enterprise and mission-critical environments. This article investigates a proof-of-concept infrastructure where blockchain technology is leveraged as an integrity layer to enhance the security posture of AI systems. Using the Breast Cancer Wisconsin (Diagnostic) Dataset as a high-stakes case study, we propose a hybrid architecture that records cryptographic hashes of model parameters and training metadata to create an immutable audit trail. To ensure experimental rigor and mitigate overfitting, the model was validated using a 70/30 stratified split and 5-fold cross-validation, achieving a validation accuracy of 98.00% alongside a sensitivity of 0.98 and an F1-score of 0.97. Our implementation demonstrates that anchoring this diagnostic model into a Vyper smart contract consumes approximately 110,472 gas. While this indicates that forensic integrity is technically viable, the results suggest that Layer 2 scaling is a functional requirement for high-frequency clinical environments to ensure economic sustainability. This approach enables post-hoc verification, supports forensic analysis, and provides a technical foundation for alignment with emerging AI risk management frameworks.