Emerging Scams · AI & Deepfakes
How to Spot an AI Voice Scam Before It Costs You
AI can clone a loved one’s voice from three seconds of audio. Here are the seven tells, the exact words to say on the call, and how to protect your family.
Updated Aug 2026 · 7 min read
The phone rings. It’s your daughter’s number — or close enough. She’s panicked: an accident, she needs money right now, and please don’t tell Dad. Every instinct says help. That instinct is exactly what the scammer is counting on.
Here’s the part that should stop you cold: that voice may not be your daughter at all. In 2026, criminals can clone almost anyone’s voice from as little as three seconds of audio — a clip from a TikTok, a voicemail, or a birthday video. The tools are cheap, some are free, and the scripts are ruthless.
Why 2026 is different
The FTC logged a reported 1,200% increase in deepfake-related complaints between 2023 and 2025, and the FBI’s 2025 Internet Crime Report flagged voice-cloning fraud as a top emerging threat. The AARP estimates one in five Americans over 65 has been targeted, with reported average losses between roughly $11,000 and $18,000. The takeaway: you can no longer trust a voice on its own. Verification, not recognition, is your defense.
The 7 signs you’re on a call with an AI voice
Manufactured urgency
Real emergencies are chaotic; scams are scripted to rush you. Demands to act in minutes — wire money, buy gift cards, read a code — are the biggest red flag. Urgency is the weapon.
A request for secrecy
“Don’t tell Mom.” “Keep this between us.” Isolation stops you from doing the one thing that breaks the scam: checking with someone else.
Odd emotional flatness
AI voices often have a flat range, unnaturally even pacing, or strange emphasis on names and numbers — eerily smooth even while claiming to panic.
Breathing that doesn’t fit
Newer clones add breaths, but they land in the wrong spots or repeat identically. Real breathing ties to emotion and sentence length.
A sterile or looping background
Genuine calls carry the acoustics of a real place. Many synthetic clips are eerily clean, or have background noise that loops and repeats.
Missing human micro-sounds
Listen for what’s absent: lip smacks, throat clears, the tiny imperfections of a real mouth. AI also repeats a word with pixel-perfect, identical intonation.
It can’t handle the unexpected
A clone follows a script. Ask a specific personal question only the real person could answer — not something on social media — and a scammer dodges, stalls, or gets it wrong.
Create your family safe word
The single best defense. One secret word your family memorizes — a clone can’t answer it. Generate one and print a card in two minutes.
Generate a safe word →Already gave money or information? Act fast
- Call your bank’s real fraud line immediately — within the first 60 minutes, some transfers can still be stopped.
- Report it to the FTC (reportfraud.ftc.gov) and the FBI’s IC3 (ic3.gov), ideally within 24 hours.
- Document everything: the number, the time, what was said, and any receipts.
- Warn your circle — scammers often work through one family’s contacts.
Frequently asked questions
How much audio does someone need to clone a voice?
As little as three seconds of clear speech can be enough for modern tools to produce a convincing clone.
Can I tell a cloned voice just by listening?
Sometimes — using the tells above — but don’t rely on it. Most people can’t reliably distinguish a good clone from a real voice. Always verify through a second channel.
What’s the single best protection?
A family safe word, combined with the habit of hanging up and calling back on a known number. Together they defeat almost every version of this scam.
Who should I report an AI voice scam to?
In the U.S., the FTC (reportfraud.ftc.gov) and the FBI’s IC3 (ic3.gov). Report even if you didn’t lose money — it helps investigators.
Sources: FBI 2025 Internet Crime Report; U.S. Federal Trade Commission; AARP fraud research. Figures are drawn from public reporting and should be verified against primary sources.