Deepfake Video Calls and Fake Celebrity Ads: How to Spot Them Before You Pay

Seeing is believing — until it is not. AI can now imitate a person’s face and voice inside a live video call, and it can put a celebrity’s face into an advert for a product they have never heard of. The FBI’s Internet Crime Complaint Center (IC3) warns that criminals use AI-generated video in real-time chats impersonating company executives, law enforcement and other authority figures, and use AI images of celebrities and social personas to promote counterfeit products and non-delivery schemes. This guide shows what to look for — and the verification habits that work even when your eyes cannot. It is part of our complete guide to AI scams and deepfake protection.

Two kinds of fake, one goal

The fake video call. You join a call and see a face you recognise — a boss, an official, a “lawyer”, a relative. The face moves, the voice matches, and the request is urgent: transfer money, share a code, keep it quiet. According to the FBI’s December 2024 public service announcement on generative AI fraud, criminals use AI-generated video in exactly this way — real-time video chats with alleged company executives, law enforcement or other authority figures. The person on screen may be a live performer whose face and voice are being re-rendered by software in real time.

The fake endorsement. An advert shows a famous face apparently recommending an investment, a gadget or a health product. Sometimes a still image carries the lie — the IC3 notes AI-generated images of celebrities and social personas promoting counterfeit products or non-delivery schemes — and sometimes a fabricated video does, with the celebrity’s voice cloned over a script they never read. Either way the economics are identical: borrowed trust, converted quickly into a payment, before the platform or the celebrity’s lawyers take the advert down.

What to look for: the FBI’s imperfection checklist

The FBI advises staying alert for subtle imperfections in AI-generated images and video. Its published list is worth saving: distorted hands and feet; irregular or unrealistic facial features; unrealistic accessories such as glasses that change shape; teeth and eyes that look wrong — too smooth, too still, or blinking unnaturally; inaccurate shadows; watermarks; lag time between the person moving and the image responding; voice not quite matching the face; and unrealistic movements, such as a head that never quite turns naturally.

On a call, add live tests of your own. Ask the person to turn their head fully sideways, wave a hand slowly in front of their face, or hold up an object you name. Current real-time systems struggle with fast occlusion and profile views. Change the subject to something personal and unscripted — a shared memory, a family detail — and watch for hesitation the technology cannot fill. And listen: the FBI notes that the tone and choice of words in a message, not just the sound of the voice, can reveal a fraud attempt. A “chief executive” who has never spoken to you suddenly asking for urgency and secrecy is out of character in a way no rendering can hide. Our guide to why AI images look wrong explains the visual glitches — hands, reflections, lighting — behind several of these tells in more depth.

The checks that work when your eyes cannot

Imperfections are a shrinking advantage: each generation of the technology fixes the previous one’s glitches. The durable defence is procedural, and the FBI’s own tips are built on it. Verify identity out of band: hang up, research the organisation’s genuine contact details yourself, and call the phone number directly. Refuse the channel’s rules: a real employer, bank or police department can reach you through official channels and will not insist the conversation stay inside one surprise video call. Never share sensitive information with people you have only met online or over the phone, and do not send money, gift cards, cryptocurrency or other assets to people you do not know or have met only online or by phone.

For celebrity endorsements, the routine is commercial rather than technical. Check whether the celebrity’s verified accounts mention the product — genuine endorsements are campaigns, repeated everywhere, not one advert on one page. Search the product name together with the word “scam” or “complaint”, and buy only through the brand’s official store, never through the advert’s own link. If the famous face is selling an investment, treat the advert itself as the warning: regulated investments are not marketed through synthetic celebrity videos. If money has already gone, our guide on your first 24 hours after being scammed lays out the recovery steps in order.

What this means

Deepfake video attacks something deeper than eyesight: it attacks the social shortcuts humans run on. For a century, a moving image of a person speaking was treated as near-conclusive evidence — news footage, court exhibits, a face on a call. When that evidence becomes cheap to fabricate, the burden of proof quietly moves. The viewer, not the video, now has to do the verifying. That is an uncomfortable inversion, but it is also a familiar one. Photography went through it when retouching arrived; documents went through it when desktop publishing made perfect letterheads free. Society adapted by trusting channels and provenance over appearances — an official filing, a verified account, a call placed to a number you found yourself.

The comparison matters because it tells US readers where the real protection will come from. Some of it will be technical — detection tools and content credentials are improving — but none of it will arrive fast enough to cover tonight’s video call. The FBI’s guidance is therefore deliberately low-tech: imperfections buy you suspicion, and procedure gives you safety. In practice that means households and small businesses should treat any unexpected video request involving money, codes or secrecy the way banks treat payment instructions — confirm through a second channel, every time, without embarrassment. The celebrity-ad version deserves the same discipline: in 2024, US consumers reported losing more than $12.5 billion to fraud overall, and fake endorsements feed the investment and shopping categories within that total. A famous face in an advert is no longer marketing evidence; it is simply a claim, and claims get checked. The people who keep their money in the deepfake era will not be the ones with the sharpest eyes. They will be the ones who made verification a habit before they ever needed it.

Small businesses should read the FBI’s executive-impersonation warning as addressed to them personally. A bookkeeper who receives a video call from “the owner” authorising an urgent payment is facing the business version of the family-emergency call — and the defence is the same shape: a standing rule that payment instructions arriving by any single channel, however lifelike, are confirmed through a second one. Write the rule down before the call comes, because in the moment, the face on the screen will argue otherwise.

Bottom line

Seeing is no longer believing — the viewer has to do the verifying now. Slow down, verify through a second channel, and treat urgency itself as the loudest alarm in the room.

Read Next

• AI Scams and Deepfake Protection: The Complete Guide
• Voice Clone Scams: How They Work and the Family Code Word That Stops Them

Sources

• FBI Internet Crime Complaint Center — Criminals Use Generative Artificial Intelligence to Facilitate Financial Fraud (PSA I-120324-PSA, 3 December 2024) (accessed 7 November 2026)
• US Federal Trade Commission — New FTC Data Show a Big Jump in Reported Losses to Fraud to $12.5 Billion in 2024 (accessed 7 November 2026)