"To Catch a Deepfake, You Need to Be One": Studying and Testing the Invisible Threat

The talk presents a practical research and testing of video and audio deepfake technologies in the context of online job interviews. The main objective is to demonstrate how deepfakes work in real time, what vulnerabilities they have, how these vulnerabilities can be identified using testing techniques, and, consequently, how to counter this threat.

The talk examines the architectures of neural network models used to generate deepfake, as well as tools for their automatic detection. The analysis uses a Red Team/Blue Team pentesting approach: the Red Team explores deepfakes from an attacker’s perspective, testing their functionality and limitations, while the Blue Team evaluates detector reliability and develops protection approaches based on the identified limitations.

The target audience includes QA specialists, penetration testers, HR professionals, and hiring managers who want to better understand the nature of deepfake-related threats and learn how to detect and prevent it.