Abstract
Insider threats represent critical cybersecurity challenges in modern enterprises, originating from authenticated users with legitimate access privileges. Traditional static controls are insufficient to detect subtle behavioral deviations. This paper presents IGNISYL, an AI-powered insider threat detection system integrating hybrid anomaly detection (Isolation Forest + Autoencoder + XGBoost), contextual risk scoring, and adaptive firewall enforcement. We evaluate on synthetic enterprise data with realistic behavioral overlap, achieving 95.6% ensemble accuracy with 94.1% precision and 0.4% false positive rate. The system demonstrates competitive performance relative to traditional ML classifiers, while prioritizing low false positive rates for operational efficiency. The prototype achieves sub-500ms response time, enabling real-time threat neutralization with interpretable risk factors. Results are reported on synthetic enterprise data and represent prototype-level performance.