Synthetic Biometrics for Next-Generation Identity Systems: A Comprehensive Review

The accelerated development of the generative AI has greatly changed the environment of the biometric identity systems. In the last ten years, Generative Adversarial Networks (GANs), Recurrent Neural Networks (RNNs), diffusion models, and transformers have made it possible to synthesize very realistic biometric identifiers, such as face, fingerprint, voice, and handwritten signatures. The developments present effective ways to address the long-term issues of the lack of data, demographic disequilibrium, and the weak regulation around sensitive identity data. Meanwhile, they also present new threats, such as deep fake-based spoofing, template inversion, adversarial manipulation, and systemic amplification of bias. This general survey discusses the history of synthetic biometric generation, threat models, forensic detection tools, as well as privacy-sensitive models. It brings together the insights of biometric security, forensic analysis and privacy-conscious AI systems, mentioning methods to perform reversible de-identification, detect deepfakes in the frequency domain, multimodal authentication, and optimisation-based model improvement. The review also compares new methods such as fairness-conscious evaluation procedures, decentralized identity systems, and machine unlearning in identity revocation and regulatory adherence. On the basis of this synthesis, an enhanced research agenda has been stated to inform the safe, fair, and confidential creation of the next-generation synthetic biometric identity systems.