A Multi-Level Strategy for Deepfake Content Moderation under EU Regulation

View on arXiv ← Back to list

Authors: Max-Paul Förster, Luca Deck, Raimund Weidlich, Niklas Kühl

Published: 2025-07-10 08:08:42+00:00

AI Summary

This paper analyzes existing deepfake detection and labeling methods for their effectiveness under EU regulations. Finding individual methods insufficient, it proposes a multi-level strategy combining technical and trusted detection methods with a scoring mechanism for scalable and practical deepfake content moderation.

Abstract

The growing availability and use of deepfake technologies increases risks for democratic societies, e.g., for political communication on online platforms. The EU has responded with transparency obligations for providers and deployers of Artificial Intelligence (AI) systems and online platforms. This includes marking deepfakes during generation and labeling deepfakes when they are shared. However, the lack of industry and enforcement standards poses an ongoing challenge. Through a multivocal literature review, we summarize methods for marking, detecting, and labeling deepfakes and assess their effectiveness under EU regulation. Our results indicate that individual methods fail to meet regulatory and practical requirements. Therefore, we propose a multi-level strategy combining the strengths of existing methods. To account for the masses of content on online platforms, our multi-level strategy provides scalability and practicality via a simple scoring mechanism. At the same time, it is agnostic to types of deepfake technology and allows for context-specific risk weighting.

Key findings

Individual deepfake detection methods are inadequate for meeting regulatory requirements. The proposed multi-level strategy offers scalability and practicality through a scoring system, addressing limitations of existing methods. This approach is agnostic to deepfake technology types and allows for context-specific risk weighting.

Approach

The authors propose a multi-level strategy combining marker-based checks (watermarks, etc.) with a multimodal approach integrating technical (artefact-based and undirected methods) and trusted (expert and crowd-sourced) detection methods. A scoring system aggregates results to determine appropriate labeling.

Datasets

Dataset-1: 1,594 papers from Scopus, arXiv, Beck, and Juris databases published between 01.01.2019 and 01.01.2025, filtered to 54 papers.

Model(s)

The paper analyzes various existing deepfake detection models but doesn't propose or use a novel model itself. The mentioned models include artefact-based, classification-based (supervised deep learning), and anomaly detection models.

Author countries

Germany

← Previous