17 July 2026

That video call was not him: spotting a deepfake before you pay or sign

Ricard runs a building renovation firm with three employees. One Wednesday morning he gets a Teams call from the "site manager" of an important client: a developer they have been negotiating with for months over a full renovation of three flats. On screen is the usual face, the usual voice, even the usual office background. He is asked to approve a budget increase in writing and pay a 30% deposit that same day, because "if the materials are not secured, the price is lost".

Ricard makes the transfer. Two days later, the real site manager calls him asking about a renovation he knows nothing about.

Until recently, a cloned voice was all you needed to run a scam like this. Now the technology has gone a step further: the image can be faked live as well. This is what is known as a video deepfake, and it has already reached business video calls, not just viral clips of celebrities.

How faking a video call is even possible

Current tools can take a few public photos or videos of a person (a company website, a LinkedIn profile, an interview) and generate a synthetic face that moves and speaks in real time during a call. It is not perfect, but on a mediocre connection, in poor light or on a small screen, it can pass for real if you are not paying close attention.

Combined with voice cloning, which has been around for a while, the result is a "complete" person on the other side of the screen who looks, sounds and moves like somebody you know. The goal is usually the same as always: an urgent payment, a change of bank details, or your signature on an important document.

The warning signs

Video deepfakes still have weak points. If you see two or three of these at once, be suspicious:

Odd movements of the face or mouth. Blinking that does not fit, a mouth not quite in sync with the sound, or an expression that "freezes" for a moment.

Unnatural lighting or edges around the head. There is often a slightly different contour between the face and the hair, or light that does not match the rest of the room.

Technical excuses for not turning the camera on properly or changing angle. "My connection is poor", "the camera is playing up today". If you ask them to turn or come closer and there is always a reason not to, be alert.

Urgency and financial pressure, exactly as in cloned-voice calls: it has to be decided now, there is no time to check with anyone.

A request that comes through an unusual channel: you are asked to approve a payment or sign a document over a video call when those things normally happen by email or in person.

The verification protocol

You do not need to be able to spot a deepfake by eye to protect yourself. You need one simple rule, known to the whole team:

No payment, change of bank details or important signature is confirmed just because you "saw and heard" it on a video call. It is always verified through a second channel that you initiate: you call the number you already had saved, you write to the usual email address, or you wait for a face-to-face meeting or a confirmation through a different medium from the one the request arrived on.

A simple and effective trick during the call itself: ask the person to do something spontaneous and unplanned, such as turning their head fully to the side, partly covering their face with a hand, or saying a specific word you choose on the spot. Live deepfake systems still often fail on unrehearsed movements like these.

And if it has already happened?

If you have made a payment and suspect you have been tricked, call your bank immediately to try to stop or reverse the transfer; the first minutes are what count. Report it to the police. Warn the real company or person who was impersonated, because they are probably attacking more of their clients. And tell your team openly: as with any scam, silence only helps it happen again.

AI is not the problem; the lack of verification is

The same technology that lets someone fake a face on a video call is the one that, applied well, saves many small companies hours of work every week. The problem is not AI itself, but accepting something as true just because it "looks real" on screen. With a clear second-channel verification rule, this scam stops working, however good the deepfake is.