In the modern era of decentralized workforces and cloud-native architectures, the traditional perimeter-based security model has collapsed, giving way to identity as the new primary security boundary. Identity and Access Management (IAM) systems are now the gatekeepers of enterprise resources, yet they face an unprecedented volume of sophisticated attacks, ranging from credential stuffing to advanced social engineering. This review examines the paradigm shift toward AI-Powered Identity and Access Management Systems. By integrating Machine Learning (ML) and Deep Learning (DL) algorithms, modern IAM frameworks have transitioned from static, rule-based engines to dynamic, risk-aware ecosystems. These systems leverage User and Entity Behavior Analytics (UEBA) to establish granular baselines of normal activity, allowing for the real-time detection of anomalies that signal compromised credentials or insider threats. This article categorizes current AI methodologies, including the use of neural networks for biometric authentication and reinforcement learning for adaptive access control policies. We explore how AI mitigates \"entitlement creep\" and automates the complex lifecycle of identity governance. Furthermore, the review addresses the integration of AI within Zero Trust Architectures (ZTA), where continuous authentication replaces the \"authenticate once, access forever\" model. By synthesizing recent research and industrial deployments, this paper provides a strategic roadmap for the next generation of identity security. The findings suggest that while AI significantly enhances the precision of access decisions, its success depends on data privacy, model transparency, and resilience against adversarial manipulation.