DEFENDING THE METAVERSE: A SURVEY ON DEEPFAKE DETECTION AND AVATAR-BASED THREAT MITIGATION

The proliferation of deepfake technologies, powered by generative adversarial networks (GANs), diffusion models, and transformer-based architectures, has led to a significant escalation in identity manipulation and misinformation. In virtual environments like the Metaverse—where interaction is synchronous, embodied, and immersive—the threat posed by deepfakes expands beyond traditional media falsification. This survey explores the evolution of deepfake techniques and their convergence with avatar-based deception, voice cloning, and synthetic behavioral modeling in virtual reality (VR) and extended reality (XR) platforms. It reviews state-of-the-art detection methods, including spatial-temporal models and multimodal attention-based architectures, and evaluates their applicability to the Metaverse’s unique challenges. The paper also analyzes the 2D-FACT model as a foundational architecture, examines new threat vectors such as real-time avatar impersonation and AI-driven NPC manipulation, and identifies critical research gaps in detection generalization, real-time operation, explainability, and digital identity governance. In conclusion, the paper highlights the need for a next-generation deepfake detection paradigm tailored for the interactive, multimodal, and decentralized nature of the Metaverse. Future research directions emphasize avatar-aware forensics, explainable AI, privacy-preserving detection, and trust frameworks for virtual spaces.