EUVIMEDEUVIMED entdecken
Crossref

A Multi-Layered Framework for Robust Real-Time Fraud Detection Against Deepfake Attacks in Digital Banking

Michele Trifiletti

Real-World AI Systems · 2025

DOI: 10.64797/rwas.v1i2.107 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Real-World AI Systems, 2025. Verfasst von Michele Trifiletti.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The rapid evolution of generative artificial intelligence has significantly transformed the security risks faced by digital banking systems, particularly through the emergence of highly realistic synthetic audio and video capable of enabling identity fraud. These deepfake-driven threats challenge traditional biometric authentication and liveness detection mechanisms, creating critical vulnerabilities in real-time financial environments where detection systems must operate under strict latency and regulatory constraints. This paper presents a multi-layered, system-oriented framework for real-time fraud detection that reconceptualizes deepfake detection as a robustness problem rather than a standalone classification task. The proposed architecture combines physiological signal analysis using remote photoplethysmography, continuous behavioral biometrics for session-level trust evaluation, and adversarially trained neural models designed to detect artifacts associated with generative systems. The framework explicitly incorporates regulatory considerations relevant to high-risk artificial intelligence applications, including explainability, data minimization, and edge-based processing. By integrating technical resilience with practical deployment and governance requirements, this work offers a deployable model for strengthening digital banking security against emerging synthetic media attacks. The proposed framework is evaluated through a robustness-oriented analytical approach, focusing on system-level resilience rather than isolated detection accuracy. Evaluation dimensions include attack resistance, detection latency, and cross-layer consistency under adversarial conditions. The results indicate that combining heterogeneous signals reduces single-point failure risks and increases attacker complexity. Although the study does not report benchmark-based empirical validation, the proposed evaluation logic provides a structured basis for assessing deployability in adversarial, latency-constrained, and regulation-intensive financial settings. These findings support the feasibility of robustness-driven, system-level fraud detection strategies for digital banking applications.

Zitieren

APA

Trifiletti, M. (2025). A Multi-Layered Framework for Robust Real-Time Fraud Detection Against Deepfake Attacks in Digital Banking. Real-World AI Systems. https://doi.org/10.64797/rwas.v1i2.107

Vancouver

Trifiletti M. A Multi-Layered Framework for Robust Real-Time Fraud Detection Against Deepfake Attacks in Digital Banking. Real-World AI Systems. 2025. doi:10.64797/rwas.v1i2.107

BibTeX
@article{Trifiletti2025,
  author = {Trifiletti, Michele},
  title = {A Multi-Layered Framework for Robust Real-Time Fraud Detection Against Deepfake Attacks in Digital Banking},
  journal = {Real-World AI Systems},
  year = {2025},
  doi = {10.64797/rwas.v1i2.107}
}
Kostenlos starten →