Fraunhofer develops real-time deepfake detector for…

In 2025, a Singaporean CFO fell victim to a sophisticated fraud scheme after being convinced to transfer US$500,000 during a Zoom call where every participant, including his boss, was fabricated. This incident underscores the growing threat of AI-generated scams targeting high-value corporate executives.

What happened

The attack utilized deepfake technology to simulate a meeting involving multiple participants, effectively impersonating the CFO’s colleagues to facilitate the fraudulent financial transfer. Following the rise in such incidents—Singapore alone recorded over 1,200 deepfake-related cases in January 2026 compared to only 43 the previous year—the Fraunhofer Institute for Secure Information Technology (SIT) has developed a prototype for real-time deepfake detection.

Videoconferencing presents significant challenges for traditional security methods, as compressed data streams, background noise, and blur filters often obscure the visual artifacts necessary for detection. To address this, the Fraunhofer team trained their AI model on thousands of real and spoofed recordings to simulate typical conference conditions. Their system runs locally on a high-performance laptop, analyzing both audio and video to generate a probability score when suspicious activity is detected. This score is intended to prompt users to verify identities through independent channels. While currently in the proof-of-concept stage, the researchers aim to integrate this technology into platforms like Zoom and Teams. The project is part of the ATHENE cybersecurity research center, funded by the German Federal Ministry of Education and Research and the state of Hesse. As deepfakes continue to threaten critical security processes and bank identification, experts emphasize that technical solutions must be paired with strict verification procedures, such as always confirming financial transfers through a separate, independent communication channel.