AI voice cloning fraud at Italy's top bank and why instruction checks are no longer enough

Fraudsters using AI to impersonate senior executives stole €95 million ($108 million) from Fideuram, the private banking arm of Intesa Sanpaolo, according to Reuters citing two sources. More than half has been recovered. About €36 million has not.

The transfers were approved after two apparent confirmations, a WhatsApp message and a phone call. Both answered the same question, which is who is asking, and AI has made the evidence for that question inexpensive to fake. A check on where the money is going does not depend on anyone's judgment of a voice, which makes it the layer worth strengthening.

How the Fideuram fraud worked

Reuters reports that the scheme began in February. Paolo Molesini, then chairman of Fideuram, received what appeared to be a WhatsApp message from Intesa Sanpaolo CEO Carlo Messina asking for urgent help with an overseas transaction. A call appearing to come from a senior partner at a prominent law firm confirmed the request, and the sources said the callers used AI to replicate the lawyer's voice. Molesini instructed the finance department to arrange transfers to foreign accounts, mainly in China and Hong Kong. Fideuram detected irregularities quickly and alerted banks and authorities in several countries. Intesa Sanpaolo and Fideuram declined to comment, and the sources said no Fideuram executive is under investigation.

This is deepfake CEO fraud, in which attackers impersonate an executive to obtain a payment. AI voice cloning fraud adds a synthetic voice, so the instruction arrives with the sound of someone the target trusts.

Recognition is the control AI makes cheap

Call-backs, approval hierarchies and message confirmations mostly establish who is instructing. A second confirmation adds assurance only when it comes from a different party on a different channel. In this case the party behind both channels was the same, so the second step added nothing independent.

The result is authorised push payment (APP) fraud at institutional scale, where someone inside the organisation approves the payment after being deceived. Payer authentication and payee verification are separate layers. The reported events at Fideuram involve defeating the first, and nothing in the report describes a check on the second.

Related reading on the same gap: What Malaysia's RM1.16m payment error tells us about payee verification.

What a pre-payment check on the payee adds

Recovery limited the loss. Reuters reports that roughly €53 million came back through cooperation between authorities in China, Portugal and Italy. That route depends on speed, on cooperation across borders and on the funds staying where they were sent. The sources said the remaining funds passed through a network of overseas accounts and were converted into cryptocurrencies, which weakens that third condition.

Payee verification is a pre-payment check that compares the beneficiary name and account details in an instruction with bank-held account data before funds are released. It tests the destination and does not depend on who gave the instruction. EU Verification of Payee covers euro transfers within the SEPA scheme, while the transfers in this case went to China and Hong Kong, so verification that reaches beyond a single scheme matters.

No single control stops a determined impersonation. Layers that fail independently do. A first-time beneficiary that cannot be confirmed should hold the payment for review by someone independent of the requester, whoever appears to have approved it.

Verification that does not rest on recognition

The Fideuram case supports a narrow proposition. When voice, name and message can all be manufactured, the check that holds is the one that tests where the money is going. The same applies as payment instructions come from software agents that have no voice to recognise. IPID

IPID provides payee verification for banks, payment service providers (PSPs) and corporates, using a Know Your Payee (KYP) check on the beneficiary account before a payment is released. It is built as a fraud prevention and payment integrity layer that sits ahead of the payment, independent of the channel the instruction came through.

Teams assessing where their approval controls rely on recognition can book an IPID demo. Finance leaders building the internal case can start with How to build the business case for payee verification.