The grandparent scam is decades old: a panicked late-night call from a 'grandchild' in jail or in the hospital, a lawyer who takes the phone, and a demand for quiet, urgent payment. It used to rely on bad audio and panic to cover the voice not matching.
It doesn't need to anymore. Consumer AI tools can clone a voice from a few seconds of audio — a TikTok, an Instagram story, a voicemail greeting. The voice on the phone can now genuinely sound like your son.
Why sound is no longer proof
Voice used to be biometric: if it sounded like them, it was them. That assumption is dead, and the scammers know it before most families do. The elderly are targeted hardest, but voice-clone fraud has hit CEOs, finance teams, and parents of college kids.
The pattern to watch for hasn't changed: panic + secrecy + money. An emergency too urgent to verify, a plea not to tell anyone, and a payment method that can't be reversed — wire, gift cards, crypto, or a courier sent to your door.
The defense: verification, not detection
You will not out-listen an AI clone. Don't try. End the incoming call and start a new one using a number you already saved or obtained from an official source. Contact another trusted family member if needed. That known-channel call-back — not the voice quality — is the deciding check before you send money or share information.
A family code word can be a weak supplementary signal, but it is not proof of identity. A word can leak, be learned from someone else, or be guessed. Use it only alongside a known-channel call-back, and never let the code word alone authorize payment or disclosure.
And one rule with no exceptions: if someone is coming to your home to collect cash, call 911. Real bail, real hospitals, and real lawyers do not send couriers for envelopes of money.
Sources and further help
Reviewed
