A Cloned Voice and a Fake Kidnapping: The FBI Just Put a $893 Million Number on AI Fraud

A mother heard her daughter beg for help. The daughter was never on the line.

A Houston-area woman, Deborah Del Mastro, answered her phone and heard a man say he had kidnapped her daughter. Then she heard her daughter's voice, panicked and pleading. The man wanted 20,000 dollars. Believing what she was hearing, Del Mastro wired 5,000 dollars before she learned that her daughter was safe and the voice on the line had been generated by software. As she told the Houston station KPRC, the money was never recovered.

Her daughter's voice was a clone, assembled by AI from a few seconds of audio. On June 2, 2026, the FBI issued a fresh warning about exactly this kind of scam. It arrived the same season the bureau put a number on the wider problem for the first time, and the number is large enough to change how you think about a phone call from someone you love.

What the FBI is reporting

In its 2025 Internet Crime Report, released on April 6, 2026, the FBI's Internet Crime Complaint Center (IC3, the bureau's clearing house for online fraud reports) did something it had not done in nearly 25 years. It broke out artificial intelligence as its own category of crime. That category alone drew 22,364 complaints and close to 893 million dollars in reported losses. It sits inside a far larger total: cyber-enabled crime cost Americans nearly 21 billion dollars in 2025, and people over 60 reported about 7.7 billion of that, up 37 percent from the year before.

The mechanism behind the voice scams is almost mundane, which is what makes it dangerous. A voice clone (synthetic speech that copies a specific person) no longer needs a studio or hours of tape. A few seconds of audio pulled from a social media video, a voicemail greeting, or a posted clip is enough for current tools to reproduce the cadence and timbre of one particular person. The attacker then places a call that arrives as a frantic emergency, a kidnapping, an arrest, a crash. The cloned voice supplies the one thing your skepticism was leaning on, the sound of someone you know, and the demand is always urgent and always routed through money that is hard to claw back: wire transfers, gift cards, payment apps, or cryptocurrency.

• The AI fraud category covers more than voice. The FBI also cites fake social media profiles, synthetic identity documents, and convincing video of public figures and loved ones, all used to apply pressure.
• Total IC3 complaints reached 1,008,597 in 2025, up from 859,532 the year before.
• The bureau compressed its core advice into two words, "Take a Beat," meaning resist the urgency and verify before any money moves.

Why this should worry you, not just retirees

The losses are real, and they cluster where defenses are thinnest. Del Mastro is one of thousands, and Americans over 60 absorbed 7.7 billion dollars in fraud last year, the group least likely to suspect that a familiar voice can be manufactured. Reading this as a problem for retirees, though, misses where it is going. The technique that fakes a daughter's voice fakes a chief executive's just as easily, and your finance team or your help desk runs on the same instinct Del Mastro trusted: if the voice is right, the request is probably real. Any process in your organization that authenticates a person by how they sound, a callback to "confirm" a wire, a verbal password reset, a spoken approval for a payment, now rests on a signal an attacker can forge with a few seconds of public audio. The deeper shift is the one the FBI just made official. By giving AI its own line in a report that has tracked internet crime since the dial-up era, the bureau is treating synthetic media as a standing category of fraud rather than a novelty, and conceding that the oldest authentication humans have, recognizing a voice, no longer counts as evidence of anything.

What to do about it

1. Hang up and call back on a number you already have. A real emergency survives a 30-second callback and a scam does not, because the attacker controls the inbound call but not your contact's actual phone.
2. Agree on a family or team code word now. Ask for it on any urgent money request, because a cloned voice can mimic how someone sounds but cannot know a secret that was never spoken on a recording.
3. Take voice out of your approvals and resets. Replace "confirm by voice" steps for payments and password resets with a channel the caller cannot spoof, such as a callback to a registered number or a second factor, because once a voice can be forged in seconds, hearing the right one proves nothing.

Bottom line

The scam that took 5,000 dollars from Deborah Del Mastro did not exploit a software flaw. It exploited the assumption that a voice you recognize belongs to the person you think is speaking. That assumption held for the entire history of the telephone, and AI has quietly retired it. The FBI's 893 million dollar figure is the first official measure of what that costs, and it will not be the last. Carry one question into your next security review, and into your next family dinner: when a familiar voice asks for money in a hurry, what do you check before you believe it?