Essay

Your Voice Was the Last Thing You Could Trust on a Call. It's Gone.

Brian and Wendy VanDoeselaar wired $66,026.92 to a stranger because the man on the phone sounded exactly like their loan officer. Voice cloning has not just given scammers a new trick — it has removed the one instinct people trusted most.

  • AI audio
  • Voice cloning
  • Fraud
  • Trust
A flat vector editorial illustration on a deep navy background, in the site's navy, warm cream and single-orange palette. An old-fashioned telephone handset sits toward the lower centre, its coiled cord trailing to the lower right and dissolving into loose orange particles and pixels. The top half of the frame is open navy space. No text or lettering appears anywhere.

On May 19 this year, Brian and Wendy VanDoeselaar were days from closing on a $460,000 condo in Hudsonville, Michigan. After 35 years in a four-bedroom ranch they were trading the yard work for an open kitchen and a sunroom that opened onto a deck. Then an email arrived that looked like it came from their lender. To keep the closing on schedule, it said, the couple needed to wire $66,026.92 to a named bank account within 24 hours. The phone number in the email differed from their loan officer's by only the last two digits. When Brian called to confirm the instructions, a voice answered that sounded exactly like the loan officer he had been talking to for months. The couple wired the money. The day before closing, the real loan officer sent them a statement of what they owed, and they understood. Their closing was cancelled hours before they were due to sign. “Interestingly enough, we never got confirmation that the money was received,” Brian told CNN. “It was just silence.”

The voice was the detail that sold the lie. A wrong email address could be caught — the one the scammers used actually had two extra letters in the domain, UnitedsMortgages instead of UnitedMortgage — but nobody screens the sound of a person they trust. Our ears treat a familiar voice the way we treat a familiar face, as something a body produces, not something an algorithm can assemble. In May 2026, at a real estate closing where tens of thousands of dollars were about to move, that was the exact assumption the scam was built on.

Be honest about what voice cloning can do

Voice cloning is genuinely good, and we should say so. A few seconds of someone speaking is enough. The tools that do it are cheap — CertifID's Tyler Adams, whose company worked with the VanDoeselaars to try to recover the money, told CNN scammers can clone a voice “for free or like $5” — and they can be run by someone with no technical skill. This is a real milestone in synthetic audio. It is also, for the same reason, a disaster for the only reliability check most people had left.

Voice-cloning is the scariest, because when you get a phone call from someone that … sounds just like the person (saved) in your phone, it's really hard to slow down and not trust it.

— Tyler Adams, co-founder of CertifID, to CNN

The saved contact is the telling detail. A number in your phone is a label a stranger can borrow, and has been for years — caller ID spoofing is ancient. The voice used to be the thing that survived the spoof. You could doubt the number, but the sound of your daughter, or your loan officer, or the parent you have heard a thousand times is not a number; it felt like the person. So the design of this scam was to borrow an address that was almost theirs, borrow a number that was almost theirs, and then supply the one thing no one prepared you to question, which is the sound of the person's voice.

Why the ear is the wrong tool for this job

There is a reason a voice worked as proof for so long, and it is not that voices are hard to imitate. It is that nobody ever needed to guard against this particular imitation. For nearly all of human history, hearing a familiar voice was a reliable signal, because you could not have heard a convincing copy of a specific person's voice: the machinery to make one simply did not exist. So the human ear developed no defence against it. We are strong at telling two neighbours apart and weak at telling a person from a recent forgery of that person, because the second category did not exist until about 2019. That is not a complaint about the quality of voice cloning. It is a statement about the available senses: the recogniser was never built for the threat.

A photograph stopped being proof on this site last week for exactly the same reason — the detector is downstream of a forgery that is already good. Audio is now doing the same thing, and it hurts more, because a voice is more intimate than a picture. A photograph of you is something someone might find; your voice is something the people who love you carry around unquestioned. When they hear it they stop and listen, and that is precisely the moment of maximum vulnerability.

An old scam wearing a new voice

None of this is new as fraud. The call from a desperate relative, the urgent request from a boss, the grandparent scam — all are decades old, and all worked the same human reflex. It is fair to say voice cloning did not invent distrust; it upgraded a grift that was already running. But the upgrade is the difference that matters. The old versions were a performance: the actor had to sound plausibly like someone in crisis, and plenty of victims were saved by a performance that was slightly off. Voice cloning removes the actor. The limit is no longer the scammer's talent; it is the supply of audio of the person being imitated, and people publish that constantly — social media clips, voicemail greetings, video interviews. The VanDoeselaars' case is a fair picture of the new floor: an email and a number that were almost right, and a voice that was perfectly right.

Real estate fraud cost US victims more than $275 million in 2025, a 58.5% increase over 2024, according to the FBI — the scale of the wire-fraud problem that AI voice cloning is making easier, as reported by CNN in the VanDoeselaars' case.

Source: CNN, citing the FBI (15 September 2026)

Notice what made the request believable, too. The amount was not round — $66,026.92 — because a round number is a tell in a world where no legitimate invoice arrives with a reassuring zero. That precision is its own kind of evidence. Scammers in 2026 spend real effort getting the boring details right, because they know the emotional details can now be forged well enough to be relied on. The two extra letters in the email domain were the only error in the whole chain, and it was buried in the noise of a closing.

The uselessly simple fix

The regulators moved on this before the shocking stories finished arriving. In February 2024 the FCC ruled unanimously that calls made with AI-generated voices are “artificial” under the Telephone Consumer Protection Act — which makes an unauthorized robocall using a cloned voice illegal and gives state attorneys general a clear target. The FTC's consumer guidance is blunter, and it tells you how thin our defences have become: if you get a call like this, hang up and call the person back on a phone number you know is theirs, and verify the story. A codeword agreed on in advance does the same job. These are the official countermeasures, and they are all about refusing to trust the very thing the scam relies on, because we can no longer tell.

Bad actors are using AI-generated voices in unsolicited robocalls to extort vulnerable family members, imitate celebrities, and misinform voters. We're putting the fraudsters behind these robocalls on notice.

— Jessica Rosenworcel, FCC Chairwoman, announcing the February 2024 ruling

It is worth being clear what the law does and does not do. A ruling that makes a cloned-voice robocall illegal is real, but it is downstream: by the time the wire is cut, the damage is done, and the recovery window is brutal. In this case the money moved on, and Adams said most funds diverted by criminals are nearly impossible to recover after 48 hours. So the individual protection ends up the same shape as the photography fix from last week: stop trusting the medium, and put your trust in something you can verify outside it. With pictures that is a chain of custody. With a phone it is the slightly absurd, slightly rude act of calling the person back on a number you know and asking for a word you agreed on in person — when the last thing you want to do with a frantic call from your own child is hang up on it.

Two essays in as many weeks, and the same sentence keeps arriving: trust is moving out of the content and into the process around it. A photograph no longer proves anything on its own; a voice no longer proves who is speaking. The uncomfortable cost is that the safest reflex we have left — verify out of band, be willing to be suspicious of a person who sounds exactly like the one you love — is also the rudest one. That is the real price of the convenience we were sold. The machine can now sound exactly like your family, and nobody is going to hand that back. So the question stops being “did it sound like them?” and becomes “can I check this without relying on my ears?” For the VanDoeselaars the answer to the first question was yes, and it cost them the keys to the home they were days away from buying.

Sources and further reading

We use AI tools in our research and drafting on this site. Every judgement and every recommendation is a person's.

See how the tools actually stack up:

Best AI Audio Tools →