AI for Life
How to spot AI scams: cloned voices, deepfake videos and fake messages
The voice of your son asking for money after an accident. A video of a celebrity recommending an investment. A message from your bank that looks perfect. Generative AI makes these scams more convincing, but they still rely on the same levers: urgency, emotion and a payment that is hard to reverse. Here is how to spot them, and what to do if you are targeted.
How AI changes old scams
The scams themselves are not new: fake relatives in trouble, fake advisers, fake investments. What changes is the disguise. The FTC explains that a scammer can take a short audio clip of a loved one's voice from content posted online and use voice-cloning software to make the call sound like them. The FBI lists other uses: fake profile photos and ID documents, fake celebrity endorsements, AI-generated audio to impersonate relatives or to access bank accounts, and videos used in real-time chats to pose as company executives or other authority figures.
The FBI also warns that AI-generated content can be difficult to identify. So the best defence is not to become an expert in pixels, but to check the story through another channel.
Cloned voices: the call from a loved one in trouble
The pattern is familiar: a relative or a boss calls, sounds panicked, needs money right now and asks you to keep it quiet. The FTC's advice is short: do not trust the voice. Hang up and call the person back on a number you know is theirs. If you cannot reach them, contact them through another family member or a friend.
The FBI suggests listening closely to tone and word choice, but a good clone can sound almost identical. Calling back is the check that works.
Agree on a family code word
The FBI recommends creating a secret word or phrase with your family to verify their identity. Choose something that cannot be found online (not a pet's name you posted), share it in person, and agree on the rule: in any urgent call asking for money, ask for the word. No word, no money, whatever the voice sounds like.
Explain it to everyone, including older relatives and teenagers who post videos online.
Deepfake videos and images
Official guides list visual clues that can give a fake away:
- distorted hands or feet, unrealistic facial features, odd shadows or light (FBI);
- blurred or pixelated areas, unnatural eye movements, lips that do not match the words (CNIL);
- content that is surprising or spectacular: check the source before sharing it (CNIL).
These clues are useful, but their absence proves nothing. If a video asks you to invest, pay or share data, go to the official site or account of the person or company yourself, never through the link in the video.
Fake sites and messages
AI also helps scammers write clean messages without the spelling mistakes that used to betray them. Judge the request, not the style. Warning signs include a payment by wire transfer, cryptocurrency or gift cards (FTC), a request for sensitive information from someone you only know online or by phone (FBI), and pressure to act before you can check. The FBI's advice: do not send money, gift cards or cryptocurrency to people you do not know or have met only online or over the phone.
Reduce what can be copied
The FBI suggests limiting public images and recordings of your voice, making social media accounts private and limiting followers to people you know. The CNIL adds cleaning up old photos and videos and blurring faces when you share images. You cannot erase everything, but you make the scammer's job harder.
What to do if you are targeted
- Stop the exchange. Hang up, do not click, do not pay. Then check through a channel you already know.
- If you have paid, contact your bank or payment provider immediately.
- Keep the evidence: numbers, messages, links, amounts, dates. The FBI asks for this kind of information in reports.
- Report it. In the United States: ReportFraud.ftc.gov (FTC) and ic3.gov (FBI). In France: 17cyber.gouv.fr, a public service run with the police, the gendarmerie and Cybermalveillance.gouv.fr, available 24 hours a day, which offers a diagnosis, advice and, in serious cases, a chat with a police officer or gendarme. The CNIL also points to PHAROS for seriously illegal content, and to each platform's own reporting tools for a fake video.
What to keep in mind
- No detector replaces a callback. Clues change as the tools improve; checking with the real person through a known channel does not.
- Anyone can be fooled. These scams play on emotion. Talking about them in advance, at home and at work, is the best protection.
- Report even if you did not pay. Reports help the authorities spot new schemes and warn others.
Questions
Can a scammer really copy my voice?
Yes. The FTC warns that scammers can use a short audio clip of someone's voice, taken from content posted online, and voice-cloning software to imitate a family member on the phone.
Is there a reliable way to detect a deepfake?
No single test. The FBI notes it can be difficult to identify AI-generated content. Visual clues help, but the safest check is to contact the person yourself through a channel you already know.
I already paid. Is it too late?
Act quickly: call your bank or payment provider to try to stop or reverse the payment, keep all evidence, then report the scam (ReportFraud.ftc.gov or ic3.gov in the United States, 17cyber.gouv.fr in France).
Sources
- FBI, Internet Crime Complaint Center, Criminals Use Generative Artificial Intelligence to Facilitate Financial Fraud, PSA I-120324-PSA (2024)
- Federal Trade Commission, Scammers use AI to enhance their family emergency schemes (2023)
- CNIL, Hypertrucage (deepfake) : comment se protéger et signaler les contenus illicites ? (2026)
- Cybermalveillance.gouv.fr, 17Cyber, le nouveau réflexe d'assistance en ligne pour toutes les victimes de cybermalveillance (2024)
One AI email a week. Only what changes something for you.