Fraunhofer develops real-time deepfake detector for video calls

As AI‑generated fraud rises, and video‑based scams seek out high-value targets, the fightback is afoot. The Fraunhofer Institute for Secure Information Technology (SIT) in Darmstadt, Germany, has come up with a prototype.

Deepfaked live-video scams are increasing. Singapore alone recorded more than 1,200 cases in January 2026, up from just 43 a year earlier. One high‑profile incident in 2025 saw a Singaporean CFO tricked into transferring US$500,000 during a deepfake Zoom call in which every participant, including his boss, was fabricated.

Fraunhofer’s prototype combines audio and video analysis to flag suspicious activity during live meetings. The system runs locally on a high‑performance laptop, avoiding the need to send sensitive footage to external servers. When the software detects multiple warning signs, a probability score indicates the call may be manipulated. This can prompt users to verify identities through additional questions or a separate communication channel.

Researchers say videoconferencing poses multiple challenges. Compressed data streams, fluctuating quality, blur filters and background noise can obscure the visual artifacts traditional detection methods rely on. The team trained its AI model using thousands of real and spoofed recordings, simulating typical videoconference conditions to improve accuracy.

The technology is still in the proof‑of‑concept stage. Fraunhofer plans to work with companies and videoconferencing providers to integrate the system into platforms such as Teams or Zoom, or into secure corporate server environments. Legal questions, such as whether participants must consent to analysis, remain under review.

Deepfakes could compromise bank video identification, and other critical security processes, researchers warn. The advice is for technical detection to be paired with clear procedures. If a financial transfer is requested during a call, a second independent channel should always be used for confirmation.

SIT’s project, under the Media Security and IT Forensics department, was created as part of “ATHENE,” a research center for cybersecurity funded by the German Federal Ministry of Education and Research and the state of Hesse.

Real-time deepfake detection for videoconferencing is among the emerging technologies examined in Biometric Update’s 2026 Deepfake Fraud Detection Market Report, which analyzes the technologies, standards, vendors and market outlook shaping AI-enabled identity defense.

Article Topics

AI fraud  |  biometric liveness detection  |  deepfake detection  |  deepfakes  |  Fraunhofer Institute IGD  |  research and development  |  video conferencing