Unmasking Synthetic Realities in Generative AI: A Comprehensive Review of Adversarially Robust Deepfake Detection Systems

View on arXiv ← Back to list

Authors: Naseem Khan, Tuan Nguyen, Amine Bermak, Issa Khalil

Published: 2025-07-24 22:05:52+00:00

AI Summary

This systematic review analyzes state-of-the-art deepfake detection methods, revealing a significant lack of evaluation for adversarial robustness across image, video, audio, and text modalities. The authors contribute a GitHub repository aggregating open-source implementations to facilitate further research and testing.

Abstract

The rapid advancement of Generative Artificial Intelligence has fueled deepfake proliferation-synthetic media encompassing fully generated content and subtly edited authentic material-posing challenges to digital security, misinformation mitigation, and identity preservation. This systematic review evaluates state-of-the-art deepfake detection methodologies, emphasizing reproducible implementations for transparency and validation. We delineate two core paradigms: (1) detection of fully synthetic media leveraging statistical anomalies and hierarchical feature extraction, and (2) localization of manipulated regions within authentic content employing multi-modal cues such as visual artifacts and temporal inconsistencies. These approaches, spanning uni-modal and multi-modal frameworks, demonstrate notable precision and adaptability in controlled settings, effectively identifying manipulations through advanced learning techniques and cross-modal fusion. However, comprehensive assessment reveals insufficient evaluation of adversarial robustness across both paradigms. Current methods exhibit vulnerability to adversarial perturbations-subtle alterations designed to evade detection-undermining reliability in real-world adversarial contexts. This gap highlights critical disconnect between methodological development and evolving threat landscapes. To address this, we contribute a curated GitHub repository aggregating open-source implementations, enabling replication and testing. Our findings emphasize urgent need for future work prioritizing adversarial resilience, advocating scalable, modality-agnostic architectures capable of withstanding sophisticated manipulations. This review synthesizes strengths and shortcomings of contemporary deepfake detection while charting paths toward robust trustworthy systems.

Key findings

Current deepfake detection methods show high accuracy in controlled settings but lack robustness against adversarial attacks. The insufficient evaluation of adversarial robustness across all modalities is a major limitation. A GitHub repository is provided to promote reproducibility and further research.

Approach

The review categorizes deepfake detection into two paradigms: detecting fully synthetic media and localizing manipulated regions. It analyzes existing uni-modal and multi-modal approaches, emphasizing the need for improved adversarial robustness and cross-domain generalization.

Datasets

CelebA, FFHQ, ForenSynths, LSUN, DiffusionForensics, CNNSpot, ProGAN, COCO, ImageNet, LAION, GANGen-Detection, UnivFakeDetect, MSCOCO, FF++, Celeb-DF, DFDC, WildDeepfake, FakeAVCeleb, DeeperForensics, ForgeryNet, DFD, Seq-DeepFake, KODF-LS, LSR+W2L, DF40, ASVspoof 2015, ASVspoof 2019, ASVspoof 2021, In-the-Wild, FakeAVCeleb, CVoiceFake, SONICS, WaveFake, EVDA, CLEAR, VSA, GRID, CD-ADD, ISOT, TweepFake, OpenLLMText, PHEME, FA-KES, WebText, Ch-9, RealNews, Enhanced TweepFake, SynSciPass, D3, CDDB-Hard, FakeClass, FakeClue, FakeQA, DGM4, StyleGAN2, Latent Diffusion, Flickr, SD2, SD3, SDXL, DALL-E, Twitter, AVLips, DeepfakeTIMIT, LRS2, LRS3, KoDF, pDFDC, AV-Deepfake1M, LAV-DF, VidTIMIT, ASVS2015, ASVS2021LA, ASVS2021DF, MUSIC-21, DF-TIMIT, FakeOrReal, InTheWild, DefakeAVMiT, RAFV, CASIA, NIST16, Columbia, Coverage, HiFi-IFDL, Places365, Dolos, WildRF, CollabDif, DMID, IID-74K, DEFACTO, MMTD, RTM, Multi-attack, FFA-VQA, GRIP, VideoSham, HTVD, FMLD, SORA, Vimeo-90K, ADD2023, PartialSpoof, VoxPopuli, LibriSpeech, Expresso, Half-truth

Model(s)

Various CNNs, Transformers, 3D CNNs, BERT, CLIP, Wav2Vec, RawNet2, RawNet3, Res-TSSDNet, LCNN, MesoNet, XceptionNet, Swin-Small, WideResNet, DeiT-S, StegaStamp, RivaGAN, GROVER, BERT-Defense

Author countries

Qatar

← Previous