I called my father last month and made him pick a word. He is an electrical engineer who spent forty years being the calmest person in any room, and he still humored me for about a minute before asking why. Because, I told him, if I ever call you crying at two in the morning asking for money, the voice on the line might not be me. Pick a word only we know, and if the person calling cannot say it, hang up. He picked one. I am not going to tell you what it is, which is the whole point.
I review gadgets for a living, and I usually open by telling you how long I lived with a thing before I formed an opinion. This is different. The product here is a crime, and the thing I want you to buy is a habit that costs nothing. So let me disclose my method up front: I did not talk to any victims, because I do not do original reporting and I am not going to stage one for effect. Everything below traces to the FBI's own crime report and the Federal Trade Commission's own advice pages. I will show you where each number comes from.
Cheap for the criminal, worst for seniors
Start with the scale, because it is new. The FBI's Internet Crime Complaint Center, IC3, released its 2025 report this April, and for the first time in the center's roughly twenty-five-year history it broke out artificial intelligence as its own crime category. That category logged 22,364 complaints and nearly $893 million in losses across all ages. People over 60 accounted for $352 million of that, from more than 3,100 complaints that specifically named AI. The FBI is blunt that this undercounts the real total, because most victims never realized AI was in the room at all. You do not file a complaint about a technology you did not know was used on you.
Here is why the older number is the one that should worry a family. In 2025, Americans 60 and older filed 201,266 fraud complaints and reported losing $7.75 billion, a 59 percent jump over the year before. The average loss for a senior victim was $38,500, which the FBI notes is nearly double the overall average of about $20,700 across every age group. More than 12,000 older people lost over $100,000 each. When a scam lands on a retiree, it takes roughly twice as much, because that is where the savings are.
Now translate the technology into a plain question: what does it actually cost a criminal to sound like your grandchild? Almost nothing. Voice-cloning tools can build a convincing copy from about three seconds of public audio, one birthday video, one graduation clip, one voicemail greeting. Security researchers at Group-IB, reported by Biometric Update in January, put dark-web prices around $30 for a one-off voice clone and roughly $5 for a synthetic-identity kit that bundles one in. A step up from that, CNN reported in May that full fraud kits rent by subscription for about $60 a month. To see how far the floor dropped: a minute of bespoke deepfake video fetched up to $20,000 as recently as the year before that January report. The price of impersonating your family fell off a cliff, and it kept falling.
A code word beats the clone, since it can't fake knowledge
The defense does not have to keep up with any of that, because it attacks the one thing the clone cannot fake, which is knowledge. A synthetic voice is trained on how your grandson sounds, not on what he knows. It works from a script. Ask it something only the real person could answer and the whole illusion collapses.
That is exactly what the FTC keeps telling people, in language that has barely changed since 2023. Do not trust the voice. Call the person back on a number you already have, not one the caller gives you and not one you find by searching. If you cannot reach them, try another family member. Ask a question a stranger could not have: the agency suggests asking something like what kind of dog you have, or where the family spent Thanksgiving last year. Security experts add one more layer: agree on a family code word in advance, something that would never surface in a social media post, that you never write down in a text or an email. A cloned voice can copy the panic. It cannot answer the dog.
Two more tells worth memorizing. Manufactured urgency is the tactic, not the emergency: a real crisis is still a crisis two minutes later, after you hang up and dial back. And the payment method is the confession. No hospital, no police station, no bail bondsman on earth takes gift cards, wire transfers, or cryptocurrency. If the voice steers you there, it is not your grandchild.
One last figure, and it is the reason none of this feels as urgent as it should. The FTC has cited research that only about 5 percent of mass-market fraud victims ever report it to a government agency or the Better Business Bureau. Separately, the agency has estimated that the true annual toll of fraud runs near $196 billion, against the $12.5 billion consumers actually reported for 2024. The losses you read about are the small, visible tip of something much larger, held down mostly by embarrassment.
So here is my recommendation, the same shape as any review I write: do the low-tech thing, and do it this week. Set a code word with the people you would wire money to in a panic. Agree, out loud, that either of you can hang up and call back and neither will take offense. It is free, it takes one phone call, and it defeats a $60-a-month machine. The accessory the industry will try to sell you next is an AI service to detect the AI voice. Skip it. You already own the countermeasure. It is a question about a dog.
