Federal prosecutors charge that one fraud scheme, which ran from July 2024 to February 2026, stole more than $129 million from companies worldwide, and the list of victims is a random cross section of the American economy. An investment bank in New York, hospital systems in Missouri and California, a bank in Rhode Island, a mutual fund in Massachusetts, a fintech company in the United Kingdom, and a pet food company in Florida. They had almost nothing in common except one shared condition: an employee who was convinced that a trusted executive needed help moving money, and then moved it.
The common thread was not a broken firewall. It was a person told that a senior executive needed discreet help with a confidential transaction, and who then wired money to accounts on the other side of the world. Prosecutors describe the mechanics as follows, and the order matters.
What the prosecution alleges
According to an indictment filed in the Southern District of New York, the scheme ran from July 2024 to February 2026 and was carried out by three foreign nationals. Prosecutors say the group used artificial intelligence to impersonate the voices and images of corporate executives and attorneys at major law firms, targeting employees who were typically in accounting departments.
The pitch, per the U.S. Attorney's office, was to tell those employees that a confidential financial transaction needed a discreet hand, then steer them toward attorneys at large international law firms who were also working on the deal. Those attorneys were, in reality, part of the scheme. One of the conspirators would then instruct the employee to wire large sums to bank accounts set up around the world, after which the group converted the money into cryptocurrency to conceal the source, ownership, and control of the stolen proceeds.
The suspects have since been taken into custody. Jacob Soall, who reportedly went by the names Rocky Balboa and Seba Clarke, was arrested in Spain in June and allegedly used a script titled Lawyer Call to sound like a real attorney. Sharon Hazan, accused of laundering the proceeds, has a prior 2008 bank-fraud conviction in this district. A third suspect, Guy Sheetrit, was arrested in the United Kingdom. Each of the seven charges, which range from wire fraud to aggravated identity theft, carries a maximum of twenty years in prison.
The one channel that was never locked
The first place a gap opened is also the most uncomfortable, because it is basic. The victims authenticated authority by listening to a voice. A phone call is the one channel through which someone can claim to be a CEO and only prove it by sounding like one. When an organization accepts a voice on the line as identity, it has built its verification around a claim rather than a check, and a skilled impersonator can fill that gap.
This is not a gap in speech-recognition technology. It is a gap in the policy that said the voice on the phone was enough to approve a wire. Many finance teams have password rules, card swipes, and approval thresholds. Almost none of them have a rule that says: if a payment instruction arrives by an unexpected channel, confirm it on a different one, through a number you already have.
The pressure that beats the process
The second gap is where normal safeguards might have stopped the transfer. A wire to a foreign account followed by a conversion to cryptocurrency is exactly the kind of transaction that normally triggers a second signer, a reconfirmation call, or a pause. Anyone who has seen this kind of fraud can name the tell: the money moved on a confidential instruction that should have required far more than a phone call.
The result is a pattern that operators should recognize: a confidential payment that normally asks a second set of eyes, arriving over the phone from someone directing the employee to external attorneys. A workflow that leaves a single employee as the final check on a transaction like that one is open to the same failure.
The assumed-trusted middleman
The third gap is subtler and worth pinning down. The fraud worked in part because the employees were directed to consult attorneys at large international law firms, and those attorneys were acting for the other side. In most payment workflows the outside advisor is a final independent check, the person you call when the numbers look odd. When that advisor is secretly on the fraudster's end, the last control in the chain simply vanishes, without anyone noticing that it is gone.
The lesson is that trust in a title can be as exploitable as trust in a voice. A letterhead, a firm name, a LinkedIn profile, a confidentiality stamp, and now an AI-generated image can all carry the same authority. The gap is never the prop. The gap is the absence of a process that verifies the prop independently of the person delivering it.
Where the gaps actually lie
So where do they lie, when an investment bank, two hospital systems, and a mutual fund all lose money to the same trick? They lie in the same place: the exact moment an organization accepted a verbal instruction as sufficient authority to move money, and let urgency and silence cancel out the controls built to make you pause.
The artificial intelligence did not create that gap. It made one person the suspects could not verify sound more convincing, longer, and more viscerally real. Had the victims required a separate confirmation beyond the phone call itself, that impersonation may never have turned into a processed wire.
The gaps, then, are not spread across a wide range of industries and countries. They are concentrated in a single, repeatable failure: trusting the source over the process. The defense is not a better AI. It is a rule that no payment instruction is real until it has been confirmed through a channel the organization already controls, no matter whose voice is on the other end of the line.
Sources
- 3 foreign nationals allegedly used AI to mimic CEOs in $129M scheme, DOJ says (NBC News, 2026-11-06)