Deepfake jako bezpečnostní hrozba: praktické testování a hodnocení detekčních systémů.

Abstract

This diploma thesis deals with the phenomenon of deepfakes and focuses on practical testing and evaluation of detection systems for recognizing content created using artificial intelligence. The theoretical part of the thesis deals with the field of generative artificial intelligence, the typology of deepfakes, its development and legal regulation of the issue. Furthermore, the diploma thesis discusses approaches and techniques for recognizing AI-generated content. The practical part of the thesis deals with experimental testing of detection tools. For this purpose, a custom dataset of videos was created containing downloaded authentic videos and content-identical generated synthetic videos with defined parameters, which were subsequently tested using detection tools within DeepfakeBench and the Hive commercial platform, where the performance of individual tools was evaluated using binary classification metrics.