How to detect an AI-cloned voice
Updated: August 2026 · By César Cerrudo
For years we used the voice as informal proof of identity. "I heard him — it was him."
Artificial intelligence is undermining that assumption. Today it is possible to generate increasingly convincing synthetic voices.
That's why, for important requests, a voice alone should not be enough to authenticate a person.
How a typical scam works
You get a call. You hear something like:
The combination is powerful: a familiar voice + fear + urgency. Your brain wants to solve the emergency before analyzing it.
What to do
Don't argue with the presumed scammer. Just verify.
Hang up and call that person's usual number. If they don't answer, contact someone else who can check on the situation.
Create a family word or question
Families can agree in advance on a word or procedure that isn't public. It shouldn't be something easy to find by browsing social media.
More important than a specific "secret word" is having a family protocol:
No urgent transfer happens without a second verification.
The project's Family Mode is a free game for two generations that includes this scenario and ends by generating a printable Family Digital Agreement, secret passphrase included.
Don't try to become a human audio detector
Listening for small imperfections in the voice can occasionally help, but it shouldn't be your main defense. The technology will get better.
Your verification procedure can keep working even if the imitation is perfect.
When the stakes are high, verify identity through more than one channel.
The book's artificial intelligence chapter covers deepfakes, voice cloning and AI-powered phishing.
DOWNLOAD A HACKER'S SECURITY GUIDE FREE →