Robust Cyber Threat Intelligence Analysis under Perturbed Attack Conditions

The growing sophistication of cyber threats, intensified by adversarial artificial intelligence, has made detecting perturbed inputs a critical requirement for modern Cyber Threat Intelligence (CTI) systems. Traditional rule-based, signature based and hybrid mechanisms often fail to recognise subtle manipulations in digital artefacts, allowing attackers to bypass detection. To address these limitations, this article proposes a multilayered CTI architecture that integrates internal network telemetry, open-source intelligence (OSINT), and dark-web monitoring to provide a unified threat view. A machine-learningdriven framework built on a hybrid Convolutional Neural Network and Bidirectional Long Short-Term Memory (CNN BiLSTM) model with its enhanced functions is introduced to strengthen perturbed-state detection and improve adversarial robustness. Evaluated on a balanced Kaggle dataset, the model achieves 99.2% accuracy, an Area Under the ROC-AUC of 0.998, and 97% system resilience, outperforming several existing hybrid CTI and intrusion detection approaches reported in recent literature. These results demonstrate the potential of hybrid machine-learning models to strengthen resilience against adversarial manipulations and shift CTI practices from reactive threat monitoring toward proactive intelligenceled cyber defence. Future work will focus on realworld deployment, broader perturbation categories, and governance strategies to support transparency and ethical compliance.