“I’d Know That Voice Anywhere”: Inside the €95 Million Fraud That Brought an Italian Bank to Its Knees

Between February 23 and 25, Fideuram executed eleven transfers totaling approximately ninety-five million euros.

The first reported payment sent €4.9 million to a Portuguese company called Vertentarticulada Lda, with an account at Banco BPI. Six more payments sent approximately €34.9 million to accounts at the same bank. Another €42.5 million went to the Bank of China.

Then Banco BPI raised the alarm.

The Portuguese bank flagged suspicious activity. Fideuram’s internal security mechanisms responded, and the real Lino Mainolfi contacted Molesini to find out what was happening.

Molesini called Carlo Messina.

Messina hadn’t sent the WhatsApp message.

He hadn’t requested the acquisition.

There was no deal.

Eleven transfers had already left Fideuram.

And the lawyer?

Paolo Nastasi had never made that call.

According to investigators, the familiar voice Molesini heard on the telephone had been recreated with artificial intelligence.

You know that moment in Mission: Impossible when someone reaches beneath their chin, pulls off an impossibly convincing mask, and suddenly everything you thought you knew about the preceding conversation changes?

There was nothing to peel away here.

The voice itself was the mask.

Modern AI tools can generate convincing imitations of someone’s voice from a relatively short recording. The Federal Trade Commission warned in 2023 that criminals could use the technology to imitate company executives and persuade employees to transfer money.

We don’t know where the criminals obtained Nastasi’s voice sample or which software they used. The published accounts of the investigation don’t identify either.

But someone had managed to make a respected international lawyer appear to participate in an acquisition he knew nothing about.

And Nastasi’s was only one of the identities involved. The opening message carried Messina’s name. The lawyer’s voice supplied apparent legal authority. A separate call, using Mainolfi’s identity, prepared the treasury department to expect Molesini’s instructions.

All three communications pointed toward the same nonexistent deal.

Fideuram and cooperating banks began trying to recover the money.

Approximately €42.5 million was returned through the Bank of China. Another €13.1 million was frozen in Portugal and later subjected to judicial seizure. According to the detailed accounting reported from the Italian court order, approximately €39.5 million remained unaccounted for.

Reuters separately reported approximately €53 million recovered and €36 million missing. Those figures have not been fully reconciled with the judicial-account breakdown reported in Italy.

The remaining funds had moved through additional foreign accounts toward services associated with cryptocurrency transactions. Milan prosecutors and the Carabinieri followed the trail, seeking assistance from authorities abroad.

And during the investigation, the name Paolo Nastasi appeared in another fraud.

This time, the target was a nonbanking company within the Bper group. The identity being impersonated, once again, was the lawyer’s, not the perpetrator’s.

The amount involved was much smaller than the Fideuram heist. The professional identity the criminals borrowed was familiar.

Paolo Nastasi’s name. Again. Used by someone else.

The published account doesn’t identify the people behind that second approach or establish whether the same voice recording was used.

But now investigators had encountered Nastasi’s identity in another case.

And other banks were reporting frauds with familiar elements.

At Banca Ifis, a manager had reportedly been deceived by fraudulent communications appearing to come from a colleague. She authorized transfers totaling approximately twenty-four million euros, routed through accounts in Spain, Singapore, Hong Kong, and Bahrain. Around twenty million euros was later recovered.

Another, smaller cooperative bank had been caught in a fraud involving approximately two million euros, part of which was recovered in Croatia.

Prosecutors were investigating these cases separately. They had not established that the same people were responsible for all of them.

Back at Fideuram, however, the money trail was producing leads.

Investigators traced funds through foreign accounts and cryptocurrency-related services. Approximately four million euros in proceeds were associated with a suspect described as a 48-year-old Israeli national.

The investigation hadn’t established publicly who placed the calls or prepared the documents. The suspect’s precise role remained under examination.

Then la Repubblica reported something about his past.

He had reportedly been convicted in the United States years earlier for participating in an international fraud conspiracy.

That crime also involved impersonating corporate executives to order large wire transfers.

The newspaper’s account pointed toward a case in the archives of the U.S. Department of Justice.

And the American court records were unusually detailed.

On August 16, 2018, the U.S. Department of Justice announced that Harry Meir Mimoun Amar, an Israeli-Moroccan national, had been sentenced to 43 months in federal prison.

Amar had pleaded guilty to participating in an international business email compromise conspiracy. He and his associates impersonated senior company executives to convince employees to transfer corporate funds into accounts controlled by criminals.

The supposed reason for the transfers was often a confidential financial transaction.

Sometimes, a secret corporate acquisition.

La Repubblica reported that the 2026 Fideuram suspect had been convicted in this earlier American case. That connection rests on the Italian newspaper’s reporting; the full 2026 Italian judicial order has not been independently reviewed for this article.

The U.S. case had led to seventeen arrests in early 2017. Amar was arrested in Israel and extradited to the United States. In April 2018, he pleaded guilty to conspiracy to commit wire fraud.

He admitted participating in the scheme from approximately January through August 2014, working with conspirators in other countries, including Turkey and Bulgaria.

But the document that explained how the organization worked had been filed even earlier.

In October 2016, federal prosecutors submitted a sixteen-page superseding indictment.

It described the people, equipment, research, recruitment, and international logistics allegedly required to carry out the fraud.

The scheme began in 2014.

According to the indictment, conspirators targeted employees at large companies, impersonated senior executives, and told their targets they had been entrusted with confidential transactions. They then directed company funds into accounts controlled by members of the operation.

The work was taught.

Prosecutors alleged that participants traveled to Israel to learn the fraud from people already conducting it, including Amar. He allegedly agreed to teach others in exchange for a percentage of the proceeds.

They had places to operate, too.

The indictment called them “safe houses”: locations used to conduct the fraud, periodically moved from country to country. Participants allegedly established bases in Turkey and Bulgaria, recruiting people with the language and technical skills needed to contact their targets.

The indictment describes an attempt to recruit an English-speaking participant and bring him to Turkey.

Before approaching a target, the conspirators allegedly did their homework.

They paid for access to an online database that helped them identify employees authorized to transfer large sums of company money. The same research helped them find the names and contact details of senior executives whose identities they intended to borrow.

Then came the lawyers.

The indictment described domains designed to resemble legitimate company addresses, along with entire websites for fictitious outside counsel supposedly working on the secret acquisitions.

Someone who decided to check whether the supposed lawyer existed could find a website supporting the story they’d already been told.

And the equipment?

The indictment listed text messages, online chats, WhatsApp, Wickr, virtual private networks, and voice-changing software among the methods and tools allegedly used.

Voice-changing software.

In an operation that began in 2014.

The filing doesn’t establish exactly how the voice-changing software was used or how convincing it was. Nor should it be confused with the generative AI tools available today.

But the organization already had a method for finding financial decision-makers, borrowing executive authority, creating fictional lawyers, and preparing international payments.

The older conspirators allegedly built an entire environment around the lie before asking someone to act on it.

Amar’s guilty plea established his participation in the broader conspiracy. The particular training arrangements, safe houses, and other operational details remain allegations in the 2016 indictment.

Nothing in the available records establishes that the 2014 organization survived intact into 2026, or that its former members committed the other frauds being investigated in Milan.

What the records do establish is how much of this apparent twenty-first-century crime had been practiced years earlier.

The secret acquisition. The executive’s instructions. The apparent lawyer. The research into employees with transfer authority. The offshore accounts.

And now, with generative AI, criminals had access to a new way of impersonating someone a target already knew.

The mask had become easier to make.

The business of wearing one was already well established.

It’s tempting to read this story and conclude that executives need to become better at recognizing impostors.

But that places an extraordinary burden on people doing jobs that already require them to make complicated decisions, maintain relationships, and respond to urgent requests.

The public reporting doesn’t fully explain Fideuram’s internal approval controls, so we cannot say which particular safeguards were absent or bypassed.

It does show where the deception ultimately succeeded: a series of high-value transfers was executed for a transaction the parent company’s real chief executive had never requested.

That is where the protective design has to begin.

An unusual acquisition should be independently confirmed through an established corporate channel. Requests for exceptional payments need verification that originates outside the same chain of messages and calls that introduced the transaction. High-risk transfers need approval processes that cannot be satisfied solely by two apparently familiar voices agreeing with each other.

The institution can make those checks a normal part of the transaction.

The person receiving the call shouldn’t have to invent a fraud investigation in the middle of a work day.

There is a second consideration, upstream of the attempted payment.

Attacks like this can depend on assembling credible information about the people being impersonated and the people being targeted.

Names, positions, professional relationships, personal contact details, employment histories, and publicly accessible records can help an attacker understand how to approach someone and what kind of story might sound believable.

Some of that information is necessary for people to do their jobs. Executives need to be visible. Lawyers need to establish their credentials. Business relationships depend on people knowing whom they’re dealing with. But a mountain of other information is exposed simply because directories, data brokers, and other services have collected and republished it.

Organizations can examine that exposure and reduce information that serves no legitimate public or business purpose.

Doing so won’t erase a public speech or prevent every attempt at voice cloning. It can remove some of the details criminals use to research targets and construct believable approaches.

And it can be managed by the organization rather than left to individual executives to discover and clean up themselves.

That’s the combination worth pursuing: independent authorization where money and authority are involved, and less unnecessary personal information available to help criminals manufacture credibility before the first message is ever sent.

The Fideuram fraud was finally exposed when the real Carlo Messina was contacted.

His answer was simple. He had never requested the acquisition.

The call came after eleven transfers had already left the bank.