Your phone rings late at night. It is your son. He is crying. He says he has been in an accident and needs money right now. The voice is his — the same tone, the same way he says your name. But your son is asleep in the next room. What you just heard was an AI voice scam, and it is now one of the fastest-growing frauds in the world.
The frightening part is how little the criminal needs. A short clip of someone talking is enough. This guide explains how these calls work, how much money people are losing, and the one simple habit that stops the scam cold.
What is an AI voice scam?
An AI voice scam is a phone call or voice message where a criminal uses software to copy a real person’s voice. The copy — often called a “voice clone” — says whatever the scammer types. The goal is almost always the same: make you panic and send money fast.
In one line: an AI voice scam turns a few seconds of someone’s recorded speech into a fake emergency call from a person you love.
These scams go by a few names. Police often call them “family emergency scams” or “grandparent scams”, because older people are a common target. The trick itself is old. AI just made the voice believable.
How much of your voice does AI need to copy it?
Very little. Researchers at the security company McAfee tested free voice-cloning tools and found that three seconds of clear audio was enough to build a clone that matched the real voice about 85% of the time. With a handful of extra clips, the match climbed to roughly 95%.
Three seconds is nothing. It is the length of a voicemail greeting. It is one sentence in an Instagram story. McAfee’s team also found more than a dozen voice-cloning tools sitting free on the open internet, needing no special skill to use.
Where scammers find your voice
- Social media videos — Instagram Reels, TikTok, YouTube, Facebook posts where you speak on camera.
- Voicemail greetings — a scammer calls, you do not pick up, and your recorded greeting plays for them.
- Nuisance calls — someone rings and simply keeps you talking for a minute while recording.
- Work content — webinars, podcast interviews, company videos and conference talks.
How much money are people losing to AI voice scams?
A lot, and it is finally being counted. The FBI’s 2025 Internet Crime Report, published in April 2026, tracked AI-assisted fraud as its own category for the first time. It recorded 22,364 complaints and more than $893 million in losses in a single year.
That figure only covers crimes reported to the FBI’s Internet Crime Complaint Center in the United States. Most fraud is never reported at all, because victims feel embarrassed. The real number is almost certainly higher, and the same scams run in the UK, Australia, Canada, India and across Europe.
A McAfee survey of 7,000 adults found one in four had either been hit by an AI voice scam or knew someone who had. Among people who received a cloned-voice message, 77% said they lost money.
You can read the FBI’s own figures in its annual Internet Crime Reports.
Can you hear the difference between a real voice and a clone?
Usually not, and you should not try. Modern clones are good enough to fool close family. There is no single sound that proves a voice is fake, so using your ears as a lie detector is a losing game. Checking, not listening, is what protects you.
That said, some calls do carry small clues:
- Odd pauses. A short delay before each reply, because a human is typing the words.
- Flat emotion. The voice says panicked things without really sounding panicked.
- Too clean. The audio sounds smooth and studio-like, with no street noise or breathing.
- Wrong background. They claim to be at a police station, but the line is silent.
- They dodge video. Ask to switch to a video call and the excuses start.
Do not lean on any of these. A good clone has none of them.
The one habit that stops an AI voice scam: a family safe word
A safe word is a short phrase that only your household knows. If someone calls claiming to be family and asking for money, you ask for the safe word. A scammer holding a perfect copy of your daughter’s voice still cannot say a word he has never heard.
The takeaway: AI can copy a voice, but it cannot copy a secret.
UK bank Starling Bank has pushed this idea hard through its Safe Phrases campaign, and fraud teams around the world now give the same advice. Setting one up takes five minutes at dinner tonight.
How to pick a good safe word
- Make it random. “Blue elephant” is good. Your dog’s name is not.
- Never use anything findable online — birthdays, pet names, school names, your street.
- Do not put it in a WhatsApp group or an email. Say it out loud, in person.
- Tell everyone who might be copied: kids, parents, grandparents, brothers and sisters.
- Change it if you ever think someone outside the family has heard it.
What to do if you get one of these calls
The scam runs on speed and fear. Slowing down beats it almost every time.
- Do not act on the call you received. Say you will ring straight back, then hang up.
- Call the person yourself using the number already saved in your phone. Never a number the caller gives you.
- Ask for the safe word if you cannot reach them and the caller is still on the line.
- Ask something only they would know — what you ate together last Sunday, not their birthday.
- Refuse unusual payment methods. Gift cards, crypto and cash handed to a courier are scam signatures. A real emergency does not need Amazon vouchers.
- Report it. In the US, file with the FBI’s IC3. In the UK, Action Fraud. In India, the cybercrime helpline 1930. In Australia, Scamwatch.
How to make your own voice harder to copy
You cannot delete your voice from the internet, but you can shrink the supply.
- Set social accounts to private, or limit who can see videos where you speak.
- Replace a personal voicemail greeting with your phone’s default robot message.
- Do not answer unknown numbers. If you do, avoid opening with “hello, yes, this is me” — let them speak first.
- Talk to older relatives about this exact scam. They are the main target and the least likely to have heard of it.
The same “check before you trust” habit applies to other AI fakes too — including the AI-generated doctors handing out health advice on TikTok. And if you use AI tools yourself, it is worth knowing that attackers can reach your AI account without ever stealing your password.
Frequently Asked Questions
How long does AI need to clone a voice?
About three seconds of clear speech. McAfee researchers found that a three-second clip produced roughly an 85% voice match, and a few extra clips pushed it to about 95%. That is short enough to lift from a social media video or a voicemail greeting.
Is AI voice cloning illegal?
Cloning a voice to steal money is fraud everywhere, and that is what gets prosecuted. The tools themselves are legal in most countries because they have honest uses, such as audiobooks and accessibility software. Rules on consent for copying a voice are still being written in the US, the EU and elsewhere.
Can my bank tell if a caller is an AI voice?
Some banks now run software that listens for signs of fake audio, but no system catches every clone. Do not assume your bank will spot it for you. A safe word and a callback to a saved number stay the strongest protection you control.
Are older people really the main target?
They are targeted heavily, because the “grandchild in trouble” story works and older adults often have savings on hand. But anyone can be hit. Criminals also clone company bosses to trick finance staff into approving payments.
What should I do if I already sent money?
Call your bank immediately and ask them to try to recall the payment, because the first few hours matter most. Then report it to your national fraud body, such as the FBI’s IC3 in the US or the 1930 cybercrime helpline in India. Keep your call logs and payment receipts.
