Live Deepfakes in Video Calls: Verify the Request, Not the Face

A colleague appears on a video call and asks you to send money to a new account before a deadline. You recognize the face and voice, so the request feels familiar. Yet the call alone cannot establish who is making it. The US FBI has warned about criminals using virtual meetings and altered media in payment scams. A convincing picture can be part of the deception, and a plain still image with fake audio can be enough in some cases.

What can change between camera and screen

In one approach, software receives a camera feed, follows a person’s expressions and produces new frames that make another face appear to speak. Research has demonstrated real-time facial reenactment, including a method that transfers movements from one person to footage of another. Audio can be altered separately. The meeting app then carries the resulting video or sound to other participants. The method varies, and not every suspicious call uses AI: a stolen account, a photo, prerecorded video or a false name can also mislead.

The important difference from an ordinary edited clip is timing. A prerecorded clip cannot easily respond to an unexpected question, while a live operator with altered media may continue the conversation. That extra interactivity can make the meeting seem more credible. It still does not prove the person’s identity. The attacker may know enough about a workplace or family to answer routine questions, and a genuine participant can be using a poor camera or connection.

Some people look for a strange blink, blurred edge or mismatch between lips and sound. Such details may raise suspicion, but they are weak as a deciding test: a clean-looking stream may be altered, and compression or a bad connection can make a real call look odd. Asking the person to turn their head or hold up an object is also only a supporting clue. The stronger check is whether the request fits a separate, established process.

A payment request needs its own route

Suppose the caller claims to be your manager and sends new bank details in the chat. Stop the transfer. Use a known phone number or another channel you already used with that person or organization; do not call a number supplied in the same meeting or message. Follow the organization’s normal approval steps, including any independent second approver. A face on screen should not replace a payment control.

The FBI’s guidance on business email compromise describes criminals inviting employees into virtual meetings and using a still image of an executive with fake audio while claiming the video or sound is not working. Its later warning also notes generated videos in real-time chats with alleged executives. These are documented scam patterns, not proof that every remote meeting is risky. They show why a demand for urgent money or credentials deserves verification even when the caller seems familiar.

If the conversation already led to action

If you sent funds, contact the bank or payment provider immediately through a known route and ask whether the transfer can be recalled. If you shared a password or access code, change it in the real service and alert the people responsible for that account. Preserve the meeting invitation, messages and payment details for your organization or local fraud authority. Acting quickly may help, but no step guarantees recovery.

A live call is useful for conversation, but its picture and sound are still data that can be substituted before they reach you. Treat the request according to its consequences: the greater the cost of a wrong action, the more important a separate check becomes.