The FBI's latest numbers: 22,364 complaints involving artificial intelligence in 2025, with losses of nearly $893 million. One slice of that is growing faster than everything around it — scams built on a cloned voice.
The bureau's 2025 Internet Crime Report calls voice-cloning fraud a top emerging threat. The FTC saw deepfake-related complaints jump 1,200% between 2023 and 2025. On most measures, AI voice scams are the fastest-growing fraud category in the country right now.
What the call sounds like
The setup is ancient. Someone in your family is in trouble — a crash, an arrest, a kidnapping — and the money has to move right now. What's new is that the voice on the other end sounds like theirs.
Making that voice doesn't take much anymore. A short clip from a public video, a podcast you appeared on, a social post with sound on. Whatever you've already put online is usually plenty.
Deborah Del Mastro's case shows how well it works. A caller told her he'd kidnapped her daughter and wanted $20,000. What sold it wasn't the story — it was hearing what sounded like her daughter begging for help. She sent $5,000 before she realised what had happened.
Sit with that for a second, because it's why ordinary scepticism doesn't save you here. She wasn't talked into anything. She recognised a voice she'd known her entire life.
What to listen for
Cloned audio is good, not perfect. People report odd pauses, and a flatness or wobble in a voice they know well — small wrong notes that are very easy to wave away when you're scared.
The dependable warning signs aren't in the audio at all. Any demand that you act this minute, with no time to check, is a red flag no matter who you think you're talking to. Same with the payment method. Wire transfer, Venmo, gift cards, crypto — those all say the same thing: somebody wants this money to be unrecoverable.
Genuine emergencies almost never require one specific irreversible payment channel inside the hour.
Two things that actually work
The FTC's advice is refreshingly plain: hang up, then call back on a number you already have. Not one the caller gives you. Not the one on your screen. The one already sitting in your contacts. A scammer controls the call coming in. They don't control the one going out.
The second one is free. Pick a family safe word — a short, nonsense phrase that's never been posted anywhere. Anyone calling with an emergency has to say it.
Here's why that works. A voice model can copy how someone sounds because it learned from recordings of them. It can't produce a phrase it's never heard, because that phrase only exists inside your family. The clone gets the voice. It doesn't get the knowledge.
Make it arbitrary. Not the dog's name, not the street you grew up on, nothing anyone could piece together from your profiles.
Who's getting hit
Older adults are still the main target, and not because they're gullible. The grandparent-emergency framing works emotionally, they're more likely to have savings within reach, and they're less likely to have run into this particular trick before.
Which makes it a conversation worth having now, before it's needed. You cannot invent a verification protocol mid-panic — and stripping out the time to think is the entire design of the scam.
Set the safe word this week. Agree on the call-back rule. Five minutes each, and both keep working no matter how good the cloning gets.
Image: Sarah Blocksidge, via Pexels
