What happened, as reported
Corriere della Sera first reported the case. Reuters, citing two sources familiar with the matter, reported on 26 September 2026 that fraudsters “using AI to impersonate senior executives” took €95 million, about US$108 million, from Fideuram, the private-banking arm of Intesa Sanpaolo, Italy's largest lender.
The sequence, as the sources described it: in February, Fideuram's then chairman received what appeared to be a WhatsApp message from the group chief executive seeking urgent help with an overseas transaction. A phone call followed that appeared to come from a senior partner at a prominent law firm, confirming the instruction; the callers used AI technology to replicate the lawyer's voice. Believing the request genuine, the chairman instructed the finance department to arrange transfers to foreign accounts, mainly in mainland China and Hong Kong.
Fideuram detected irregularities quickly and alerted banks and authorities in several countries. Roughly €53 million was recovered with authorities in China, Portugal and Italy. Reuters's sources said “some €36 million remains missing”, converted into cryptocurrency after passing through overseas accounts. Those two figures total €89 million, not €95 million; the report does not explain the difference, and other outlets converting to dollars gave slightly different sums.
Millions of euros, according to two sources cited by Reuters, 26 September 2026
Source: Reuters, 26 September 2026, as republished by VnExpress on MSN (see Sources).
Recovered and missing sums as reported total €89 million, not €95 million; Reuters did not explain the difference.
The chairman resigned in March citing personal reasons. Neither he nor other Fideuram executives are under investigation, the sources said, while Milan prosecutors have placed a foreign national living outside Europe under investigation on suspicion of computer fraud. Intesa Sanpaolo and Fideuram declined to comment. Reuters noted an earlier Italian case in which fraudsters mimicked a government minister's voice to persuade the businessman Massimo Moratti to transfer nearly €1 million, which was recovered.
Where the cloned voice was aimed
This section is our reading of the reported facts.
The chief executive was impersonated in text, a WhatsApp message, which needs no cloning at all: a profile photo and a name are enough. The voice that was copied was the lawyer's, used in the step meant to confirm the request. The fraud was built to pass a check that a careful executive would make, by supplying a second, apparently independent voice from outside the company.
That matters for how companies think about deepfake risk. The common framing is that an executive whose voice is widely available on podcasts and video can be cloned to give orders. In this case the senior voice was not needed, and the voice that was needed belonged to an outside adviser, whose public audio a company does not control and may not know exists.
It also shows what turned the case. The bank caught irregularities in the transfers themselves and moved fast enough to recover more than half. The fraud succeeded at the human layer and was limited at the transaction layer.
What the FBI advises
The FBI's Internet Crime Complaint Center has published two public service announcements on this pattern. In December 2024 it warned that criminals use AI-generated audio “to impersonate well-known, public figures or personal relations to elicit payments”. In May 2025 it warned of a campaign, under way since April 2025, of text messages and AI-generated voice messages claiming to come from senior US officials, often with a request to move the conversation to a separate messaging platform.
Its advice is consistent: independently find a phone number for the person and call to verify; if contacted by someone you know on a new platform or number, verify through a previously confirmed channel; do not send money on the strength of a message alone; and agree a secret word or phrase. The December 2024 announcement also says, “If possible, limit online content of your image or voice”.
That last line collides with how executives build reputations. Podcast interviews, conference talks and video are the core of executive visibility, and they cannot be withdrawn once published. A company can reduce some exposure, but it cannot make a visible executive inaudible, and this case suggests it would not have mattered.
What a company should do
What follows is our reading, not security advice for any particular organisation.
Make verification independent of the channel the request came in on. A confirming call that the requester arranges, or that arrives on the requester's timetable, is not verification. A call-back to a number already on file, made by the person being asked, is. That applies to outside advisers as well as executives.
Publish the rule internally and to counterparties. State that the chief executive and directors never request payments or transfers by messaging app, and that urgent confidential transactions still go through normal approval. A rule that staff know exists is easier to invoke under pressure from a senior name.
No single person moves large sums, however senior. The chairman could instruct the finance department. Two-person approval that includes someone outside the request chain would have forced a second check.
Prepare the communications as well as the controls. The resignation came in March with no explanation; the story surfaced seven months later through the press, with the bank declining to comment. Companies should decide in advance what they will say if they are defrauded this way, because it is increasingly a matter of when. Visible executives remain worth having; the record their visibility creates, discussed in podcast and interview footprint, is also what lets the public tell a real statement from a fake one.
Methodology and limitations
This report is based on Reuters's report of 26 September 2026, which we read as republished in full by VnExpress through MSN syndication, and on TechRadar's summary of 29 September 2026, which also cites Reuters; Corriere della Sera's original report was not read. We also read the FBI Internet Crime Complaint Center's public service announcements of 3 December 2024 (“Criminals Use Generative Artificial Intelligence to Facilitate Financial Fraud”) and 15 May 2025 (“Senior US Officials Impersonated in Malicious Messaging Campaign”). All read on 5 October 2026. Four limitations apply. Every detail of the fraud comes from anonymous sources, and Intesa Sanpaolo and Fideuram declined to comment. The recovered and missing sums as reported do not reconcile with the €95 million total, and the dollar figures in other coverage differ again; we use Reuters's euro figures. Reuters does not say how the lawyer's voice was obtained or how much audio was used. And our reading of what the case means for executive visibility is ours.
Conclusion
The Fideuram fraud did not need the chief executive's voice. It used a text message in his name and a cloned voice of an outside lawyer to defeat the check meant to catch exactly that kind of message, and it cost €95 million before the bank recovered a little over half. The FBI's advice to limit public audio does not fit executives whose reputations are built on podcasts and video, and this case suggests it is not where the risk was. The defences that address it are procedural: verification through channels the requester does not control, a published rule that executives never request transfers by messaging app, two-person approval, and a communications plan ready before it is needed.
Frequently Asked Questions
How much did Fideuram lose to the deepfake fraud?
Sources told Reuters on 26 September 2026 that fraudsters took €95 million, about US$108 million. About €53 million was recovered with authorities in China, Portugal and Italy, and about €36 million remains missing; those figures as reported do not add up to €95 million.
Whose voice was cloned in the Fideuram fraud?
According to Reuters's sources, a senior partner at a prominent law firm. The group chief executive was impersonated by WhatsApp message; the cloned lawyer's voice was used in a call confirming the instruction.
When did the Fideuram fraud happen?
It began in February 2026, according to the sources. It became public in late September 2026, after Corriere della Sera and then Reuters reported it.
What does the FBI advise about AI voice impersonation?
Independently look up a number for the person and call to verify, verify any new number or platform through a previously confirmed channel, do not send money on the strength of a message, and agree a secret word or phrase. It also suggests limiting online content of your voice where possible.
Should executives stop doing podcasts because of voice cloning?
This case does not suggest so. The chief executive's voice was not used; the attackers cloned an outside lawyer and impersonated the executive by text. Procedural controls address the risk; withdrawing from public audio would not have prevented this fraud.
What controls stop this kind of fraud?
Verification through a channel the requester does not control, such as a call-back to a number already on file; a published rule that executives never request transfers by messaging app; and two-person approval for large payments, including someone outside the request chain.
Sources
- press Reuters, Fraudsters using AI to impersonate executives stole €95 million from Fideuram (as republished), 26 September 2026, read in full via VnExpress on MSN. Read 5 October 2026.
- trade press TechRadar, Italy’s top bank hit by an AI messaging scam which cost it nearly €100 million, 29 September 2026, citing Reuters. Read 5 October 2026.
- official FBI Internet Crime Complaint Center, Criminals Use Generative Artificial Intelligence to Facilitate Financial Fraud, public service announcement, 3 December 2024. Read 5 October 2026.
- official FBI Internet Crime Complaint Center, Senior US Officials Impersonated in Malicious Messaging Campaign, public service announcement, 15 May 2025. Read 5 October 2026.