The proliferation of artificial intelligence and machine learning systems across critical infrastructure has introduced novel attack vectors that traditional cybersecurity frameworks struggle to address. This research examines the implementation of AI-powered threat hunting methodologies specifically designed to detect adversarial machine learning attacks within zero-trust network architectures. Through comprehensive analysis of threat landscapes, attack taxonomies, and defensive strategies, this study presents a framework for integrating advanced detection mechanisms into zero-trust environments. Our findings demonstrate that hybrid AI-human threat hunting approaches can achieve detection rates of up to 94.7% for sophisticated adversarial attacks while maintaining acceptable false positive rates below 2.3%. The research contributes to the evolving cybersecurity paradigm by addressing the intersection of adversarial AI and zero-trust security models, providing actionable insights for enterprise security architects and threat intelligence professionals.