How voice scams are actually run
Every pattern below works the same way underneath. A voice you trust, a reason you cannot wait, and a channel the attacker chose. The voice is the part that has become cheap; the other two have always been the real mechanism.
The countermeasure, before the catalogue. Hang up and call back on a number you already have — from your own contacts, a statement, the company’s published line. Never a number the caller gives you. A clone cannot answer a phone you dialled. This defeats all eight patterns, costs two minutes, and works whether or not you can hear anything wrong with the voice.
The eight patterns
The family emergency call
A child, grandchild or sibling in trouble — an accident, an arrest, a hospital. Distress is the point: it explains why the voice sounds slightly off and why you must act now. Often followed by a “lawyer” or “officer” who takes over the call.
Needs: a public clip of the relative · your number · ninety seconds of panic
CEO and finance-team fraud
A cloned executive authorises an urgent transfer, usually alongside a plausible email thread and often timed to a real event — an acquisition, a quarter end, a holiday. The voice call exists to satisfy the second factor in your approval process.
Needs: a recorded earnings call or interview · org-chart knowledge
The bank fraud-team call
Your bank calls about suspicious activity and asks you to move money to a “safe account” or read back a code. A cloned voice may be used to reproduce a specific adviser you have spoken to before.
Needs: a leaked customer list · a spoofed caller ID
Virtual kidnapping
No one has been taken. A cloned voice crying or calling for help plays in the background while a caller demands payment and forbids you from hanging up. The instruction not to end the call is the tell, because ending the call is the entire defence.
Needs: a social video with speech · knowledge of a travel plan
Romance and long-con trust building
Voice notes rather than calls, used over weeks to make an invented person real. The cloned or synthetic voice removes the last thing that used to break these frauds — the request to actually speak.
Needs: patience · a consistent synthetic voice
Fake endorsements and investment pitches
A recognisable public figure appearing to promote a scheme, distributed as short video or audio ads. Volume rather than targeting: the clip only has to convince a small fraction of a very large audience.
Needs: any public figure · ad budget
Interview and identity fraud in hiring
A candidate whose voice in a screening call does not belong to the person who will do the job, or who uses real-time voice conversion. A growing problem in remote hiring, and one where a recorded call is usually available to check.
Needs: a remote-only process · no in-person stage
Fabricated recordings as evidence
A recording produced in a dispute — employment, custody, defamation — that shows someone saying what they did not say. The inverse also matters: a real recording dismissed as a fake because the technology now makes that denial plausible.
Needs: a dispute · a plausible chain of custody
What makes any of them work
Three ingredients, and only one of them is new.
A voice you already trust
Roughly a minute of clear speech is enough. A podcast appearance, a conference talk, a video on a family account, a voicemail. If someone has spoken publicly, assume the material exists.
A reason you cannot verify first
Every script contains an explanation for why the normal check is impossible right now — she is in custody, the deal closes in an hour, don’t tell anyone, don’t hang up. This is the load-bearing part of the fraud, not the voice.
A channel the attacker controls
An inbound call, a spoofed number, a callback number they supply, a chat thread they started. Everything holds together only while you stay inside it.
Notice that a detector attacks none of these directly. Detection tells you afterwards what a clip was. Leaving the attacker’s channel stops the fraud while it is happening — which is why the callback rule is at the top of this page and not the bottom.
Reducing the material available about you
- Agree a family phrase. One word or question, never written down or sent by message, that a caller claiming to be a relative must answer. Low technology, effective, and it costs nothing.
- Set a callback rule at work. Any payment instruction arriving by voice is verified on a number from your own directory, regardless of who appears to be calling. Make it policy so nobody has to be the person who doubted the boss.
- Shorten your public voicemail greeting. A long personalised greeting is clean, well-recorded training material that anyone who calls can collect.
- Treat urgency as the signal. Not the voice. The voice is now the cheapest thing in the attack; the pressure not to check is the expensive part.
If it has already happened
- Preserve the original recording. Do not forward it through a messaging app before it is checked — each hop re-compresses the audio and removes evidence.
- Contact your bank immediately if money moved. Recovery windows are measured in hours.
- Report it. [VERIFY: link your national fraud reporting bodies here — e.g. Action Fraud (UK), IC3 (US), Scamwatch (AU), and the relevant Indian portal. Verify each URL before publishing.]
- Tell the person whose voice was used. They are almost certainly not the only target, and they can warn the rest of the circle faster than you can.
Questions
How do I tell if a voice on the phone is AI?
Usually you cannot by listening, and trying to is the wrong move — it keeps you on the attacker’s call while you deliberate. Hang up and call back on a number you already have. Verify the channel, not the voice.
How much of someone’s voice does a scammer need?
About a minute of clear speech. Podcast audio, a conference talk, a social video or a voicemail greeting is generally enough for a likeness that survives a phone call.
Are these calls always cloned voices?
No, and it matters less than it seems. Many still use a live person and a plausible story. The countermeasure is identical either way, which is a good reason to rely on the procedure rather than on your ear.
Can Truthring tell me who made the clip?
It can often name the generation system, when the clip matches a signature we hold. It cannot identify the person who used it. That is an investigative question, not an audio one.
Reviewed