Home // SECURWARE 2025, The Nineteenth International Conference on Emerging Security Information, Systems and Technologies // View article

DeepAuthVerify - A Modular Framework for Deepfake Detection in Facial Authentication Systems

Authors:
Domenico Di Palma
Alexander Lawall
Kristina Schaaff

Keywords: Deepfake Detection; Facial Recognition; Authentication Systems; Large Language Models.

Abstract:
The rise of deepfake technologies poses a significant threat to biometric authentication systems, especially those based on facial recognition. In our study, we investigate the reliability of commercial facial recognition systems when exposed to deepfake attacks and propose a modular authentication solution (DeepAuthVerify) that integrates deepfake detection into the verification process. We developed DeepAuthVerify as a two-layered system combining deep learning-based face recognition and feature extraction with the semantic interpretability of a Large Language Model (LLM) for decision-making. Despite achieving lower accuracy (66.89%) compared to commercial solutions (OpenCV: 91.43%, Amazon Rekognition: 93.80%), DeepAuthVerify demonstrates the potential as a complementary layer for deepfake detection, enhancing transparency and modularity. The results indicate that commercial systems, when properly configured, offer robust protection against deepfake attacks. However, their black-box nature limits adaptability and auditability. Our proposed system provides a novel, extensible architecture that fosters explainability and integration into existing authentication environments. In addition to the evaluation, we publicly release the evaluation pipeline to allow reproducibility and comparability of future research.

Pages: 48 to 54

Copyright: Copyright (c) IARIA, 2025

Publication date: October 26, 2025

Published in: conference

ISSN: 2162-2116

ISBN: 978-1-68558-306-4

Location: Barcelona, Spain

Dates: from October 26, 2025 to October 30, 2025