Three second TikTok could be enough for AI scammers to use you to steal money

A Ferrari executive may have accidentally revealed one of the simplest ways to stop an AI voice scam, by asking a question the person on the other end of the phone could not possibly have researched.

In July 2024, an executive received WhatsApp messages followed by a call supposedly from Ferrari CEO Benedetto Vigna.

The caller's voice and accent had been cloned and they were requesting a transfer.

But instead of immediately complying, the executive asked the caller to name a book Vigna had recommended just days earlier. The impersonator hung up.

The incident is one example of why experts say people should stop relying on whether a voice 'sounds right' when money or sensitive information is involved.

Valeria Contreras, marketing manager at Zepo Intelligence, told UNILAD that a voice can now be cloned from as little as three seconds of audio.

Voice deepfakes rose 680% year-over-year, while deepfake fraud attempts increased 2,137% over three years, rising from 0.1% to 6.5% of all fraud attempts. In 2024, those incidents increased from roughly one a month to seven a day.

Valeria, marketing manager at Zepo Intelligence, told UNILAD that a voice can now be cloned from as little as three seconds of audio. (Getty Stock Image)

How do AI voice scams work?

The technology is already being used in several ways.

Scammers can impersonate executives during live voice or video calls to authorize transfers, imitate relatives during supposed emergencies, or begin conversations through WhatsApp or SMS before following up with a cloned-voice call.

The targets aren't limited to celebrities or senior executives, either, with Valeria saying ordinary families and mid-level employees are increasingly in scope because publicly available social media clips or a voicemail can provide enough material for a convincing imitation.

Human ability to detect deepfakes is also limited: a 2024 analysis covering 56 studies and 86,000 participants found people were only around 55% accurate at identifying them by ear or eye.

Commercial detection tools can perform much better in laboratory conditions, reaching 96% accuracy, but that can fall to between 50% and 65% in real-world conditions.

(Getty Stock Image)

How can you tell if an AI voice is a scam?

The answer may not be hidden in the sound itself.

Subtle audio artifacts or unnatural pacing can occur, but experts warn they are not reliable enough to make simply 'listening carefully' a safe strategy.

Instead, the strongest defense is independent verification. That matters because the technology is already capable of real-time interaction.

For consumers, the advice is that if a loved one suddenly calls asking for money, hang up and call them back using a known number.

If that is not possible, ask something specific and recent that would not be available on social media. Never let urgency remove the verification step.

Global fraud losses linked to generative AI are projected to rise from $12.3 billion in 2024 to $40 billion by 2027, according to the figures supplied by Zepo.

So if a voice on the phone sounds exactly like someone you trust, that may no longer be enough.

A question only the real person could answer could be the difference between a convincing call and a scam being exposed.