Covert channels in multi-tenant virtualization environments pose a substantial security threat by allowing unauthorized data transfer through shared hardware resources. This paper proposes a novel hybrid approach that integrates timing-based and storage-based covert channel techniques with dynamic rate adaptation and advanced machine learning detection. Our framework leverages real-time hardware profiling and statistical feature extraction, enabling robust covert communication while minimizing detectability. Experimental evaluation in a nested virtualization setup demonstrates throughput rates up to 24.36 bytes per second with minimal error rates, outperforming single-technique baselines. We further present a machine learning–based detection system capable of identifying covert activity with over 90% accuracy and low false positives. These findings highlight the urgency of comprehensive mitigation strategies such as traffic normalization and multi-layer monitoring to secure virtualized cloud infrastructures against covert data exfiltration.