Adversarial Deepfakes: Evaluating Vulnerability of Deepfake Detectors to Adversarial Examples

Recent advances in video manipulation techniques have made the generation of\nfake videos more accessible than ever before. Manipulated videos can fuel\ndisinformation and reduce trust in media. Therefore detection of fake videos\nhas garnered immense interest in academia and industry. Recently developed\nDeepfake detection methods rely on deep neural networks (DNNs) to distinguish\nAI-generated fake videos from real videos. In this work, we demonstrate that it\nis possible to bypass such detectors by adversarially modifying fake videos\nsynthesized using existing Deepfake generation methods. We further demonstrate\nthat our adversarial perturbations are robust to image and video compression\ncodecs, making them a real-world threat. We present pipelines in both white-box\nand black-box attack scenarios that can fool DNN based Deepfake detectors into\nclassifying fake videos as real.\n