Skip to content
← ORAVYS
Forensic intelligence // Voice fraud

He heard the minister's voice. It cost him a million euros.

In early 2025, a group of fraudsters in Italy did something that would have sounded like science fiction a few years ago. They picked up the phone, called some of the richest people in the country, and spoke to them in the voice of the Italian Defence Minister, Guido Crosetto. The voice was a clone, built by AI. The story they told was urgent and patriotic: Italian journalists had been kidnapped abroad, a ransom was needed to free them, and the Bank of Italy would reimburse every euro.

Among the targets were Giorgio Armani, the Prada family through Patrizio Bertelli, and the Beretta and Menarini families. Most of them grew suspicious and hung up. One did not. Massimo Moratti, the former owner of Inter Milan, made two transfers totalling roughly one million euros before the deception unravelled.

Before you read on: is that voicemail real?

If a recording is sitting on your phone right now and you are not sure whether it is a real person or a generated voice, you can find out before you act on it. Your first scan is free, no card needed.

Analyze a voice. Free, no card →

What actually happened

The scam worked because of one simple fact. When the man on the other end of the line sounds exactly like a public figure you have heard a hundred times on television, your brain stops checking. Recognition feels like proof. It is not. The fraudsters did not break into any account or steal a password. They borrowed something far more trusted: a familiar voice.

There is a twist worth keeping in mind, and it is the only good news in this story. The Italian police managed to trace and freeze the money. Moratti's funds were located on an account in the Netherlands and recovered. That outcome is the exception, not the rule. Most people who pay a convincing voice never see the money again, because by the time anyone realises the call was fake, the funds have already moved through several countries.

It is not only billionaires

The Crosetto case made headlines because the targets were famous and the amount was large. But the same technique is now pointed at ordinary families, and the version that frightens people most needs no minister and no million euros.

Security researchers at Bitdefender have documented what is sometimes called the crying-child scam. A parent receives a call. On the line is the voice of their own child, crying, frightened, saying they are in trouble and need money or need to be picked up right now. The voice is a clone. A few seconds of audio, pulled from a public video, a voice note, or a school recording, is enough to build it. No parent in that moment stops to run a forensic check. The panic is the whole point.

The uncomfortable conclusion

A voice used to be one of the most personal forms of proof we had. If you heard your son, your minister, or your boss, you knew who you were talking to. That assumption no longer holds. A voice is no longer proof of who is speaking. It is now just another piece of media that can be generated on demand.

So what can you actually do?

You cannot un-publish your voice, and you cannot stop a clone from being made. What you can change is whether you act on a voice blindly. Three habits help.

  1. Agree on a codeword. Pick a word or short phrase with the people who matter, family and anyone who handles money for you. It must never be spoken on a recording or typed in a chat. If a call asks for money or urgent action and the codeword is missing, treat the call as fake until proven otherwise.
  2. Slow the call down. Urgency is the scammer's main tool. Hang up and call the person back on the number you already have. A real emergency survives a callback. A scripted clone usually does not.
  3. Get a second opinion on the audio itself. If you have the recording, you do not have to trust your own ear. A neutral check can tell you whether a sample is more consistent with a real recording or with a generated one, before you decide what to do next.

Where ORAVYS fits

ORAVYS exists to give that neutral second opinion. You submit a voice sample and you get back a plain answer about whether it looks like a genuine recording or a generated one. We do not publish the inner workings of how that judgement is reached, for the same reason a bank does not publish its fraud rules. What matters to you is the verdict, delivered without you having to become an expert.

The point is not to make you paranoid about every call. It is to give you somewhere to turn in the moment that counts, when a voice you trust is asking you to do something you would normally question.

Check a voice before you trust it

Drop in a suspect recording and get a clear read on whether it is real or generated. Your first scan is free, and you do not need a card to run it.

Analyze a voice. Free, no card →

For the wider argument on what it means when a voice stops being proof, read the manifesto.

Sources cited in this article: reporting on the Crosetto voice-clone scam by Cybernews, Euronews and Identity Week (February 2025); Bitdefender HotForSecurity research on AI voice-cloning family scams. Figures and named targets are matters of public record. ORAVYS does not host or redistribute any voice sample and never reuses submitted audio without explicit consent.