The Fideuram Case: The Voice Is No Longer Proof

This article comes out days after the news broke, not out of laziness, but by choice. The case emerged on September 25, filled the financial news pages for two days, and then, as always, slipped away. Waiting for the sensationalism to settle allows us to look at the facts with greater clarity, without the rush of sensational headlines. That is what we are attempting to do here. There is a question that anyone who has ever made an online wire transfer asked themselves reading this story. To transfer two hundred euros from their own bank account, an ordinary user must pass two authentication factors, a temporary code, and sometimes a call from the bank if the amount is unusual. Yet a C-level executive, with decades of banking experience, authorized within a few hours a transfer of nearly 95 million euros based on a WhatsApp message and a phone call. It is not a naive question. It is probably the most useful question one can ask about this affair.

The Case, Reconstructed

In February 2026, the then-chairman of one of Italy's leading private banking firms—a subsidiary of one of the country's largest banking groups—received a WhatsApp message appearing to come from the parent company's CEO. The phone number was unfamiliar, but the tone and details left no doubt: a series of urgent wire transfers was required, through the subsidiary's treasury, for an overseas financial operation that the parent company could not manage directly due to practical concerns over a potential news leak. As reconstructed by the Prosecutor's Office of Milan and reported by Ansa, the chairman was not suspicious because months earlier a similar transaction had indeed collapsed precisely due to a leak regarding an international banking acquisition.

Shortly thereafter came the second move—the step that separates this case from a ordinary CEO fraud scam. The chairman received a phone call from an international lawyer he knew personally, a professional with whom he had previously worked. The voice was his, the tone was his, the colloquialisms were his. Only it was not him: it was an AI-generated voice clone built from public audio recordings of his voice, while the actual lawyer, as Il Post's reconstruction details, knew nothing about it and was never involved. Reassured by that second confirmation, the chairman authorized the fund transfers, totaling roughly 95 million euros, directed primarily to bank accounts in China and Hong Kong.

The alert system eventually triggered, and part of the stolen funds was recovered through collaboration between the Milan Prosecutor's Office and international authorities, with support from Eurojust: approximately 40 million euros were frozen in a Chinese institution and another 13 million identified in Portugal, according to the timeline established by Il Sole 24 Ore. Between 36 and 39.5 million euros remain missing depending on sources, partially converted into cryptocurrencies and now difficult to trace. Investigations led to the registration of a foreign national in the suspect registry, while the chairman involved—not under investigation—resigned in March for personal reasons that were never detailed publicly.

A Voice Is Enough, Just a Few Seconds

The technical aspect of the case is easily dismissed as science fiction, but it is in fact standard practice. Modern voice cloning tools, including open-source models capable of running offline on ordinary hardware, do not require hours of studio-quality audio to produce a convincing voice over a phone line: a few seconds of clear speech are enough—precisely the kind of material anyone who has given a conference talk, granted an interview, or participated in a webinar produces without realizing it, as explained in Pasquale Pillitteri's technical analysis of the case.

A paradox hides here that is worth highlighting. Vulnerability does not depend on how distracted or tech-illiterate a person is; it depends on how much public audio material of their voice exists. A corporate lawyer participating in panel discussions, an executive giving media interviews, a professional featured on industry podcasts—all of them, simply by practicing their profession publicly, inadvertently build the dataset from which a scammer can draw. The very authority that makes a voice trustworthy over the phone is what makes it clonable. It is not an individual security error; it is a side effect of professional visibility in a world where publishing one's voice online has become almost second nature.

Urgency and Authority, Not Naivety

Dismissing these incidents as mere gullibility misses how social engineering operates. The psychological mechanism exploited in these cases does not bypass the victim's intelligence; it bypasses their cognitive shortcuts—the same ones we all use daily to make rapid decisions without re-evaluating every situation from scratch.

Three levers operate in tandem. The first is urgency: a request that leaves no time for verification, framing every minute of delay as a concrete cost or missed opportunity. The second is authority: the request originates, or appears to originate, from the very top of the hierarchy, and questioning an order from the boss's boss requires a level of social friction that most people instinctively avoid. The third is social proof: recognizing a familiar voice provides reassurance at the exact moment suspicion ought to arise. None of these three levers require computer expertise to execute or resist; they require a process that does not rely on the instantaneous judgment of whoever answers the phone at that exact moment.

The Paradox of Responsibility

And here lies the heart of the matter, which the average person's question brings into focus better than any technical analysis. For a transfer of a few hundred euros, an ordinary account holder faces multi-factor authentication designed specifically not to trust a single confirmation channel. For operations involving vastly larger sums—those managed at the executive level—the formal level of verification is often lower, not higher, because it is assumed that the authority of the person giving the order makes verification redundant.

This is not a paradox born of executive incompetence. It is the structural result of an organizational architecture where hierarchical trust—essential for running any sizeable enterprise—becomes the exact point of failure in a world where voice identity can be replicated. The more an organization relies on top-down orders being executed swiftly, the less likely those orders are to face independent verification. The case described here is not an individual lapse of attention; it is a system failure where human verification, though present, was bypassed because it relied on a single vector—a recognized voice—that proved spoofable.

Who Is Accountable When Authority Is Fake?

The open questions raised by this case matter more than any definitive answer, and it is right to leave them open. Who truly verifies that an order comes from the person issuing it, when every communication channel can be simulated? Where in a complex organization should a verification procedure sit that does not depend on the individual answering the phone? And does liability in such cases rest on the person executing the order, on those who designed the internal controls, or on those who failed to foresee that voice cloning would become accessible to anyone?

These are not questions to dismiss lightly. It is worth remembering that in seemingly distant fields, such as autonomous software agents, the exact same underlying issue is being confronted: security cannot rely on the apparent authority of whoever issues an instruction; it must rely on independent verification separate from the sender's claimed identity. It is the same principle applied to a different domain.

What the FTC Says, and Its Limits

On the very day this story resurfaced, across the Atlantic, US Federal Trade Commission Chairman Andrew Ferguson spoke on a related topic: the liability of artificial intelligence agents. Addressing the Reuters Momentum AI conference in Austin, Ferguson rejected the idea of treating these systems as autonomous actors capable of independent initiative, declaring, as reported by Reuters, that he would continue to resist this "anthropomorphization" of tools for as long as he leads the agency. According to Ferguson, audits of logs from systems presented as "runaway" almost always reveal machines executing received instructions rather than making autonomous decisions, meaning liability should rest on those who design and instruct those systems, not on the tool itself.

We should be honest about what this statement is and what it is not. As a post-conference analysis by Forkast rightly notes, it is not a binding regulation, nor does it introduce a new legal framework specific to autonomous agents; it is a public policy stance expressed by the head of the US consumer protection authority. But it points in a clear direction, and it is telling that it arrived just as Italy was debating a case where the issue was not a runaway software agent, but a perfectly mimicked human voice. The underlying principle remains identical: if the identity of the person issuing an order can be spoofed, the only real defense is a system that does not blindly trust that identity, regardless of who or what is impersonating it.

Boring Procedures, Not Miraculous Software

The most useful takeaway from this story, from a practical standpoint, is perhaps the most uncomfortable to accept: defense is organizational, not technological, and it works better the more tedious it is to follow. No transfer above a defined threshold should proceed based on a request received solely through an informal channel, without a secondary call to a pre-registered corporate number independent of the incoming request line. A secondary authorization should be required from someone who did not receive the initial request, preventing the duped individual from being the sole validator. Transactions to previously unused beneficiaries or jurisdictions warrant a mandatory cooling-off period, however inconvenient that may feel when the request insists on urgency.

There is a rule simpler than all others, and perhaps the most effective: when a request comes from the top of the hierarchy while simultaneously demanding haste, suspicion should be triggered because of that combination, not suppressed by it. It is a counterintuitive reversal of natural instincts, which push us to trust more when the source appears authoritative. Against a familiar voice on the phone, personal intuition alone is no longer sufficient: what is needed is a process that operates independently of whoever happens to answer.

Not an Isolated Incident

Anyone assuming this affair was an isolated, movie-like anomaly should look at what is emerging around it. The same Milan Prosecutor's Office that reconstructed this case is investigating two parallel incidents using similar blueprints: a scam involving roughly 24 million euros—with funds routed through Spain, Singapore, Hong Kong, and Bahrain, partially recovered in Barcelona—and a third targeting a smaller institution for about 2 million euros, with other Italian banking groups listed among targeted entities. This is not a bespoke attack built for a single target, but a pattern being replicated with the industrial logic typical of any effective attack vector.

Statistics confirm that CEO fraud predates artificial intelligence but is being transformed in scale by it. According to FBI data reported by MilanoFinanza for 2025, the US recorded nearly 24,800 Business Email Compromise (BEC) complaints, with reported losses reaching $3.05 billion—up from $2.77 billion the previous year. A survey by the Association for Financial Professionals found that 74 percent of surveyed organizations were targeted by at least one such attempt in 2025, up from 63 percent in 2024. The chief security officer of the banking group involved in the opening story stated, in a report by Il Fatto Quotidiano, that since the start of the year alone, roughly five thousand fake AI-generated videos had been taken down, most built using the image of the group's CEO. Where setting up a credible attack against an executive once required a structured criminal enterprise, today it requires a laptop and a few minutes of public audio. It is the democratization of a technique that once demanded rare skills, and now demands mainly patience in gathering the right clues.

The Questions That Remain Open

I must return to citing an independent video game, Papers, Please, where the player's sole task is verifying documents at a border checkpoint, cross-referencing every detail against constantly shifting rules while time ticks away and errors carry real consequences. It is a game about the cognitive exhaustion of systematically doubting what seems obvious, and it is hard not to recall it when reading this story. The same tension between trust and verification runs through series like Severance, where automatic compliance with corporate procedures becomes a trap precisely because no one stops to ask who is actually giving the orders.

More questions remain on the table than answers, and it is right to leave them there. How do we genuinely authenticate the identity of someone speaking on the phone, in an era where voice is no longer sufficient proof of anything? How long will procedures deemed "boring" hold up against attacks that grow more convincing and cheaper to execute every month? And above all, must methodical suspicion toward urgent, top-down requests become a standard operating procedure for anyone managing significant funds, rather than an exception reserved for those who have already been burned?

The question that opened this article—that of the account holder amazed that they must verify more for two hundred euros than an executive did for ninety-five million—is not naive. It is probably the right question. The answer, however uncomfortable, is that this type of attack is not engineered against the intelligence of its victims; it is engineered against the way we all inevitably trust the voices we recognize. And the only defense that does not depend on who happens to answer the phone remains a slow, un-economic procedure that is annoying to follow every single time. It is the price, perhaps inevitable, of living in a time when even the voice of someone we know may no longer be theirs.