Authors:
Ghefar Alaa Aldeen Alrefai, Yao Junhui, S. J. Vimal Aravintha, S. Sharan Jeev, K. Daniel Jasper, Ahmed J. Obaid
Addresses:
Department of Informatics and Communications Engineering, EBLA Private University, Idleb, Syria. Department of Information Technology and Engineering, Jinsha Interactive, Sichuan, China. Department of Data Science, Indiana University Bloomington, Bloomington, Indiana, United States of America. Department of Cybersecurity, University of Texas at Dallas, Richardson, Texas, United States of America. School of Electrical and Electronics Engineering and Computer Science, Queens University Belfast, Belfast, Northern Ireland, United Kingdom. Faculty of Computer Science and Mathematics, University of Kufa, Kufa, Najaf Governorate, Iraq.
Deepfakes, hyperrealistic synthetic media made by generative artificial intelligence, are a major information security risk. Modern variational autoencoders, generative adversarial networks, and denoising diffusion probabilistic models can generate deepfakes that fake real-looking video, audio, and pictures. This study examines the deepfake threat to information security and digital forensics in a comprehensive, technical, and methodical way. Deconstructing significant deepfake generation frameworks includes a detailed mathematical description of the generative adversarial network minimax objective and diffusion model dynamics. Granular threat modelling ties face swap, lip sync, voice synthesis, full-body reenactment, and synthetic document generation to confidentiality, integrity, availability, authentication, and non-repudiation violations. Consider the 200,000-euro voice deepfake fraud and political misinformation. This paper presents a state-of-the-art survey of deepfake detection methods across the spatial domain (mesoscopic properties, co-occurrence matrices, and generative adversarial network fingerprints), the temporal domain (frame-to-frame optical flow and blinking inconsistencies), the spectral domain (discrete cosine transform and fast Fourier transform upsampling artefacts), and biological signal analysis. Performance measurements under adversarial conditions such as JPEG compression, H.264 encoding, and evasion show large generalisation gaps. The study concludes with a 2025–2030 research agenda, Security Operations Center deployment instructions, and policy suggestions. Academics, forensics experts, and chief information security officers will find this book valuable; it includes six original Mermaid visuals and 25 peer-reviewed references.
Keywords: Synthetic Media Forgery; Information Security; Generative Adversarial Networks; Diffusion Models; Digital Media Forensics; Adversarial Machine Learning.
Received on: 25/09/2025, Revised on: 26/12/2025, Accepted on: 25/01/2026, Published on: 11/11/2026
DOI: 10.69888/FTSCL.2026.000844
FMDB Transactions on Sustainable Computer Letters, 2026 Vol. 4 No. 4, Pages: 244-256