In February 2026, Italian private bank Fideuram, a subsidiary of Intesa Sanpaolo, fell victim to a sophisticated fraud scheme involving artificial intelligence. Scammers cloned the CEO's voice and used WhatsApp to deceive the chairman of the board, resulting in €95 million being transferred abroad. Despite the fraud being detected quickly and some of the funds being recovered, €39.5 million remains missing, with part of the money having passed through cryptocurrency wallets.
Key facts: €95 million transferred over 3 days (February 23-25, 2026); €42 million recovered from China; €13 million seized in Portugal; €39.5 million remains missing; €4 million traced through Malta, Luxembourg, the Netherlands, and Canada into Bitcoin wallets; an AI-cloned voice was used in the deception; a 48-year-old Israeli citizen is under investigation.
The fraud began on February 23, 2026, when Paolo Molesini, then chairman of Fideuram, received a WhatsApp message from someone claiming to be Carlo Messina, CEO of parent company Intesa Sanpaolo. The number was unfamiliar to Molesini, but the sender presented the request as a confidential acquisition involving an international bank. According to court documents summarized by ANSA, the alleged Messina explained to Molesini that a direct acquisition by Intesa was impossible because an information leak could prompt Italian securities regulator Consob to halt the deal. Fideuram was presented as the necessary vehicle for completing the transaction.
A second contact then appeared, supposedly from Paolo Nastasi, managing partner of A&O Shearman Italia and a lawyer Molesini knew. Nastasi had no connection to the scheme. Reports based on the investigation claim that the caller's voice was reproduced using artificial intelligence — deepfake audio technology that is becoming increasingly common in corporate fraud.
| Attack Element | Method | Objective |
|---|---|---|
| Initial contact | WhatsApp message from fake CEO | Establish urgency and confidentiality |
| Voice verification | AI-cloned lawyer's voice | Confirm legitimacy of the request |
| Documentation | Fake emails from law firm | Provide payment instructions |
| Legal documents | Confidentiality agreement, power of attorney | Create an appearance of legitimacy |
| Urgency | Need to avoid a leak to Consob | Prevent procedural verification |
Email messages designed to resemble communications from the law firm contained payment instructions and details of overseas accounts. Molesini received documents including a confidentiality agreement and an apparent special power of attorney bearing Messina's signature. This multi-layered attack used several communication channels to create an illusion of legitimacy: WhatsApp for the initial contact, an AI-cloned voice call for verification, and email for official documentation.
Ultimately, eleven transfers totaling approximately €95 million were sent abroad between February 23 and 25, with substantial amounts directed to accounts in China and Portugal. The scammers used a classic Business Email Compromise (BEC) technique but enhanced it with modern AI voice-cloning tools, making the deception significantly more convincing.
Critical success factor: The combination of social engineering (urgency and confidentiality), technical sophistication (AI voice cloning), and multi-layered verification (WhatsApp + voice + email + documents) created a convincing illusion of legitimacy that overcame ordinary verification procedures.
Fideuram detected the fraud quickly enough to stop or recover much of the original transferred amount. ANSA reported that approximately €42 million sent to China was blocked and returned, while Italian investigators secured more than €13 million held at a Portuguese bank. A Milan judge authorized the seizure in Portugal following cooperation with authorities in the country.
Court documents cited in later Italian reports put the unresolved amount at €39.5 million. At least some of this money had already moved through overseas accounts and into crypto-related channels before banks could freeze it. Investigators have since reconstructed one chain involving approximately €4 million.
Corriere reported that these funds passed through accounts in Malta, Luxembourg, and the Netherlands, then through a Canadian money transfer platform before reaching two Bitcoin wallets. This complex international route demonstrates how scammers use multiple jurisdictions and cryptocurrencies to launder money and make tracing more difficult.
| Status of Funds | Amount | Location |
|---|---|---|
| Recovered from China | €42 million | Blocked and returned |
| Seized in Portugal | €13 million | Under investigators' control |
| Missing funds | €39.5 million | Partially in cryptocurrency |
| Traced into Bitcoin | €4 million | Two wallets |
The use of cryptocurrency as a final destination reflects a growing trend in corporate fraud. Bitcoin and other cryptocurrencies offer anonymity, irreversible transactions, and the absence of central control, making them attractive to criminals. However, the public nature of blockchain also means that transactions can be traced, although identifying the owners of wallets remains a difficult task.
The difference between initial reports of €36 million being converted into cryptocurrency and later court details showing €39.5 million unresolved stems from figures disclosed at different stages of the investigation and recovery process. This highlights the complexity of tracing money in real time and how the full picture of the damage becomes clear only after the initial response has been completed.
Milan prosecutors are investigating a 48-year-old Israeli citizen as an alleged member of the group involved in the fraud. However, the individual has not been formally charged, and authorities are still verifying his identity. This reflects the typical complexity of international cybercrime investigations, where identifying and extraditing suspects can take considerable time.
Molesini himself is not under investigation. Corriere reported that Fideuram did not initiate legal proceedings against him in connection with the incident. This decision likely reflects an understanding that Molesini was the victim of a sophisticated attack rather than an accomplice to the fraud. Nevertheless, the incident still had serious corporate consequences.
Corporate consequences: Fideuram announced Molesini's departure on March 12. ANSA reported that he resigned as chairman of both Fideuram and Intesa Sanpaolo Private Banking for "personal reasons," with Tommaso Corcos taking over the chairman's duties. Although the resignation was not officially linked to the incident, the timing is evident.
This case raises important questions about corporate responsibility and verification procedures in the era of AI-powered threats. Traditional verification methods, such as calling back a known number or holding an in-person meeting, become less effective when attackers can clone voices and create convincing fake documents in real time.
The involvement of an Israeli citizen as a suspect reflects the global nature of modern cybercrime. Criminal networks often operate across multiple jurisdictions, using intermediaries and shell companies to make tracing more difficult. Israel has a highly developed technology ecosystem, which may explain why some cybercriminal operations are based there or involve Israeli citizens.
For Fideuram and Intesa Sanpaolo, the incident represents significant reputational damage even though most of the funds were recovered. Private banking clients expect the highest security standards, and a successful attack of this scale can undermine trust. The bank will likely review its verification procedures for large transfers and invest in additional safeguards against AI-powered threats.
The Fideuram incident demonstrates the evolution of corporate fraud in the age of artificial intelligence. Traditional BEC (Business Email Compromise) attacks have evolved to incorporate AI voice cloning, deepfake video, and more sophisticated social engineering. Organizations must adapt their defenses accordingly.
| Vulnerability | Traditional Protection | Required Upgrade |
|---|---|---|
| Voice verification | Voice recognition | Multi-factor verification, code words |
| Urgent requests | Standard procedures | Mandatory delay, independent verification |
| Confidential deals | Limited circle of people | Multiple approval levels |
| Documentation | Visual verification | Cryptographic verification, blockchain |
| Large transfers | Single executive signature | Multisig, time delays |
AI voice-cloning technology is becoming increasingly accessible and sophisticated. Services capable of cloning a voice from only a few seconds of audio are now available for just a few dollars, making this threat accessible even to low-level criminals. Organizations can no longer rely on voice recognition as a method of verification.
The cryptocurrency aspect of this scam also highlights the need for better monitoring of financial flows. Although cryptocurrencies are often associated with anonymity, the public nature of blockchain means transactions can be traced. Organizations should work with law enforcement agencies and crypto analytics firms to track stolen funds.
Key lesson: In the age of AI, traditional verification methods are no longer sufficient. Organizations must implement multi-layered protection combining technological solutions (AI detection, multisig), process improvements (mandatory delays, independent verification), and the human factor (training, code words).
The Fideuram incident serves as a warning to the entire financial industry. As AI technologies become increasingly sophisticated, scammers will continue adapting their methods. Successful defense requires continuously updated procedures, investment in new technologies, and a security culture in which every employee understands the risks and knows how to respond to them.
The recovery of €55 million out of the €95 million transferred shows that rapid detection and international cooperation can significantly reduce the damage. However, the €39.5 million that remains missing demonstrates that even the best systems cannot prevent every attack. The challenge for organizations is to minimize the likelihood that such attacks succeed and maximize their ability to recover when they do.
“In a world where artificial intelligence can clone any voice and create any face, trust must be verifiable, not assumed. Security is not the absence of threats, but the ability to detect them before they cause harm.”
— Bruce Schneier, cryptography and security expert
