Researchers Build a Real-Time Deepfake Detection System for Video Conferencing

Researchers have built a real-time warning system for video conferences that flags synthetic participants. It responds to deepfake technology that now imitates voice and appearance so convincingly that the standard visual check no longer works.

Why video calls became a target for fraudsters

Video conferencing has become a basic business tool. Companies use it to conduct job interviews, sign deals, and verify the identity of clients and partners. A video call has traditionally been trusted more than an email or a voice call: it seemed hard to pass yourself off as someone else while looking into a camera. That logic no longer holds.

Modern deepfake technology can overlay another person's face on a video stream in real time while cloning their voice at the same time, with a synthesis delay that has become practically unnoticeable. As a result, a call participant looks into the "eyes" of the person on screen, hears a familiar voice, and is in fact dealing with a fully synthetic image generated by a neural network.

• Deepfakes fake the face and voice simultaneously, during the call itself
• Synthesis quality has reached the point where visual analysis alone is no longer enough
• Video conferences are a direct target: negotiations, identity verification, financial decisions
• Most existing deepfake detectors work on recordings, not in real time

How the warning system works

The system analyzes the incoming video and audio stream during the call and displays a warning if it detects signs of synthesis. Its key goal is not to replace human judgment but to give people a timely signal before they make a decision based on trust in the "person" on the other end.

Deepfake detectors look for characteristic traces of generation: mismatches between facial expressions and speech, artifacts along the edges of the face and hair, anomalies in eye movement, unnatural blinking, and unusual lighting patterns. The task is complicated by the fact that generator networks keep improving, so the detector has to work in a constant technological arms race.

"More and more often we have to ask ourselves whether real eyes are looking at us from the screen," says a

TechXplore AI article on the threat of deepfakes in video conferencing.

The warning has to appear in the call interface in real time. Otherwise it loses its practical value and becomes a tool for retrospective investigation rather than live protection.

Why the solution is emerging now

By 2025–2026, deepfake quality had jumped sharply: available commercial and open-source tools make it possible to create convincing fakes without any specialized technical knowledge. The barrier to entry has dropped so far that the technology is now in the hands of a wide range of bad actors, from phone scammers to corporate spies.

AI-based deepfake detectors have existed for several years, but most of them work after the fact, analyzing video that has already been recorded. A real-time system is a fundamentally different problem: the algorithm has to deliver its verdict within milliseconds, while the conversation is still going on and before the user has made a key decision.

What this means

Practical deepfake warning tools built directly into the video call interface are an important step for corporate security and online identity verification. For businesses that use video conferencing to verify clients, hold financial negotiations, or make hiring decisions, such a tool closes a real vulnerability, one that the human eye can no longer cover in the era of generative AI.