Machine learning-driven threat detection platforms deliver high classification accuracy but persist outputs in mutable centralised storage that adversaries can falsify after compromise. Distributed ledger platforms guarantee cryptographic audit permanence but lack autonomous threat-analysis capability. Existing hybrid proposals remain confined to narrow application verticals. This paper proposes an Intelligent Cyber Protection Model (ICPM) addresses three key limitations identified in prior AI-blockchain cybersecurity systems by integrating a CNN-BiLSTM hybrid neural classifier with a Hyperledger Fabric permissioned ledger and Go Chaincode programmable contracts for confidence-stratified autonomous incident handling. Empirical validation on the CICIDS 2017 Kaggle dataset (2,830,743 records, 78 features, 14 categories) achieves 96.612% accuracy, 97.406% precision, 98.818% recall, 98.107% F1-score, and 79.066% specificity. Across 50,242 attack evaluation windows, only 594 events went undetected 1,545 fewer than a Decision Tree baseline and 111 fewer than LinearSVC. At 73,409 parameters and 0.157ms per-window inference latency, the system demonstrates suitability for high-throughput intrusion-monitoring environments. An ablation study confirms synergistic CNN and BiLSTM contributions. Blockchain scalability analysis indicates a projected 2.75× safety margin below Hyperledger Fabric’s reported PBFT throughput ceiling, while 92.7% of true positive events were assigned to the Go Chaincode ISOLATE tier.