[DigitalToday reporter Yoonseo Lee] AI voice phishing that copies a family member’s voice using just three seconds of audio to extort money is spreading rapidly across the United States.

On July 28 (local time), online outlet Gigazine reported that this type of scam is focusing on older people and that the scale of damage is growing by the day.

Sharon Brightwell, who lives in Hillsborough County, Florida, was tricked in July 2025 by a scam call that mimicked her daughter’s voice and handed over $15,000 (about 22 million won). The woman on the phone sounded like her daughter, sobbing, and said she had caused an accident by hitting a pregnant woman while driving and that police had confiscated her mobile phone. A man who then took the call said he was her daughter’s lawyer and demanded bail money. He told her not to say what the money was for when withdrawing it at the bank, explaining that it could create problems for her daughter’s credit.

Brightwell withdrew cash in less than an hour after taking the call and gave it to a delivery driver who appeared to be a court official. She later reached her actual daughter and learned that there had been no accident at all.

Such crimes are now spreading as one of the leading AI-based financial scams in the United States. The FBI’s Internet Crime Complaint Center, in its 2025 annual report released in April 2026, for the first time classified “fraud using AI” as a separate category. In 2025, it received more than 22,000 AI-related complaints and reported losses exceeding $893 million (about 1.308 trillion won). Of that, losses by victims aged 60 and older amounted to $352 million (about 515.6 billion won).

The problem is that actual damage could be larger than the statistics. The FBI said its tally is limited to cases in which victims recognized the crime and filed reports themselves. Engineer Tim Green said many AI voice phishing victims may not even know AI was involved, adding that $893 million should be seen as a lower bound, not an upper bound.

A key factor exploited by scammers is that barriers to entry for voice cloning technology have fallen sharply. With only three seconds of audio, it is possible to create synthetic speech that is hard to distinguish from the original voice, and voice data can be obtained from public content such as automated response messages, parts of podcasts and Instagram videos. A single TikTok video featuring a grandchild’s voice can provide all the information needed to carry out a crime.

The availability of cheap and numerous tools was also cited as a factor behind the spread. Consumer Reports said many major voice cloning services such as Descript, ElevenLabs, Lovo, PlayHT, Resemble AI and Speechify lack sufficient effective safeguards to prevent misuse.

Green, however, said such measures have clear limits. He said they can only help investigations after a crime occurs and victims lose money, and do little to block the stage of creating a cloned voice from three seconds of audio in the first place.

The reasons older people suffer heavier losses are also relatively clear. Older people tend to have larger savings on average, increasing the financial gains for criminals. Green said, “Even if you explain 100 times that a voice can be forged, when a voice impersonating your child or grandchild asks for help, that knowledge disappears quickly.”

As a result, discussions on responses are shifting from post-incident tracing to pre-emptive controls. As one regulatory option, Green proposed requiring identity verification at the start of using a service so it is possible to track who created a voice clone with AI. With access to AI voice technology rising, how to break the scam structure combining public voice data with low-cost tools is expected to be a key issue going forward.