Abstract
<title>Abstract</title> <p>Traditional Identity and Access Management (IAM) systems rely on static credentials and centralized authorities, leaving organizations vulnerable to single points of failure, credential theft, insider misuse, and increasingly sophisticated deepfake impersonation attacks. In this paper, we propose a Decentralized AI-powered Zero-Trust IAM (DAZT-IAM) framework that combines permissioned blockchain infrastructure, self-sovereign identity (SSI) principles, and deepfake-resistant multimodal biometric authentication (face, voice, and behavioral keystroke dynamics) with a continuous, risk-adaptive AI trust scoring engine. The proposed system is compliant with ZTA principles and checks every access request continuously, unlike the older “authenticate-once” models. Blockchain-anchored Decentralized Identifiers (DIDs) and Verifiable Credentials (VCs) eliminate the dependency on a central identity provider. The biometric pipeline includes a dedicated deepfake-detection module that employs frequency-domain artifact analysis and temporal consistency checks to counteract synthetic media spoofing. We describe the system architecture, consensus and smart-contract design, the multi-modal fusion and liveness detection pipeline, and a risk-scoring model for adaptive access decisions. The experimental evaluation on simulated and benchmark datasets demonstrates that the proposed framework provides competitive authentication accuracy, high detection rates of deepfake attacks, and low average access decision latency, while removing single points of failure for identity. Our results demonstrate that the integration of blockchain-based decentralization and AI-based continuous trust evaluation provides a pragmatic approach of resilient and privacy-preserving IAM for sustainable digital infrastructure.</p>