Guides pratiques pour vous protéger — vous, votre famille et votre entreprise — contre les arnaques liées à l'IA, les deepfakes et les nouvelles menaces cyber.
Articles populaires
The voice begging for help was cloned from a clip the family posted online
Deborah Del Mastro answered the phone and heard a man say he was holding her daughter. Then she heard her daughter, crying, pleading for help. The man demanded $20,000. Before she could think it through she sent $5,000. The money is gone, and her daughter was never in danger. The voice on that call was synthetic, generated by AI from a short recording of the real woman speaking.
Her case is one of thousands. In early June 2026 the FBI renewed its warning about AI voice cloning, the practice of using software to copy a real person's voice from a sample of audio. What makes this worth your attention is not novelty. It is how cheap, fast, and convincing the technique has become, and how cleanly it scales from a frightened parent at home to a corporate finance desk.
How the attack works
The scale is now on the record. The FBI's 2025 Internet Crime Report, released this year, carried the first dedicated section on artificial intelligence in the roughly 25 year history of the bureau's IC3 (the Internet Crime Complaint Center, where the public reports online fraud). That section counted 22,364 AI-related complaints and close to $893 million in losses for the year. The damage falls hardest on the old: Americans over 60 reported about $7.7 billion in fraud losses across all categories, up 37 percent from 2024. The bureau names three engines behind the AI surge: voice cloning, deepfake images and video, and AI-written scripts.
The voice-cloning attack itself is short and brutal. First the attacker harvests a sample of the target's voice, and the sample can be tiny. The FTC says under 30 seconds is enough, and some cases have run on roughly three seconds lifted from a social media clip, a voicemail greeting, or a posted video. That audio is fed into a cloning program that reproduces the person's timbre and cadence well enough to fool their own family. Then comes the call. A first voice plays the loved one in crisis, kidnapped, arrested, or in an accident, and in organized rings a second voice often follows, posing as a lawyer or an officer to close the deal. The script runs on two levers as old as fraud itself: urgency and secrecy. The victim is told to move money now, quietly, through channels that cannot be reversed.
- The payment demand is always something hard to claw back: a wire transfer, cryptocurrency, or gift card numbers read aloud over the phone.
- Caller ID is no defense, because the number can be spoofed to show a familiar name while the panic does the rest.
- Human ears are no longer a reliable filter. The FBI and FTC both warn that a cloned voice can now pass for the real person to the people who know them best.
Why this reaches past the family phone
The concrete cost is already counted in ruined savings and the $893 million the FBI tied to AI fraud in a single year, money that, as Del Mastro learned, almost never comes back. The reason it belongs in your next security review is that the same primitive does not care whether the target is a grandmother or a treasury analyst. Your executives' voices are not private. They sit in earnings calls, conference recordings, podcasts, and webinars, which is exactly the few seconds an attacker needs to clone a chief executive ordering an urgent payment or a vendor confirming new bank details. The shift underneath is that an old proof of identity has quietly expired. For most of human history, recognizing a voice was good enough to trust it. That assumption is now a vulnerability, and the criminals industrializing it run call centers, buy cloning tools by subscription, and target thousands of people a week. The trend points one way: the audio needed keeps shrinking, the fakes keep improving, and detection keeps losing the race.
What to do about it
- Verify through a separate channel before any money moves. Hang up and call the person back on a number you already have, because a cloned voice can carry the inbound call but cannot answer the real person's phone.
- Agree on a code word that lives nowhere online. Give your family and your finance team a private verification phrase, since the clone can copy a voice but cannot know a secret it was never able to scrape.
- Treat urgency plus an untraceable payment as the signature of the scam. Slow down when a call demands wire, crypto, or gift cards immediately, and in a business require dual authorization and an out-of-band callback for any change to payment details, because the attack only works at the speed of panic.
Bottom line
Voice cloning has crossed from a lab curiosity into a commodity crime, and the FBI's first AI chapter is the official acknowledgment that it is here at scale. The defense cannot be trusting what you hear, because that instinct is precisely what the attack is built to exploit. Carry one question into your next security meeting and one conversation at your next family dinner: if a voice you love, or a voice you answer to at work, called tomorrow demanding money in a hurry, what step in your routine would make you stop and check before the money was gone? You can read the FBI's findings in its press release on AI and cryptocurrency fraud and the FTC's guidance in its alert on AI-enhanced family emergency schemes.