<p>The emergence of generative Artificial Intelligence (AI), particularly Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs), introduces new challenges for biometric security systems. While AI research has improved methods for biometric authentication, these technologies are now the tools for producing advanced synthetic identities for defeating security measures. This paper provides a thorough survey of the current state of deepfake detection in biometric systems with regard to the evolving nature of generative models, the emerging threat of cross-modal deepfakes, and the dire need for robust multi-layer defense mechanisms. As part of the analysis, the current techniques of detection are evaluated, and their relative strengths and weaknesses are identified. Furthermore, research directions are put forward concerning advanced detection algorithms, making diverse and comprehensive datasets, and privacy-preserving technologies. The ethical use of deepfakes with respect to biometric systems further focuses on the necessity to construct new regulatory frameworks in light of accelerating generative technologies evolvement. It includes the development of sophisticated detection algorithms, multimodal datasets, and privacy-preserving methods to safeguard biometric systems against emerging deepfakes. Finally, this paper outlines future research directions.</p>