A callback to a phone number your company already had on file — not a voice-detection tool, not an AI classifier — would have stopped this scheme cold. That’s the detail worth holding onto as the headlines about AI-cloned CEO voices stealing $129 million roll out, because federal prosecutors in Manhattan just unsealed a case that proves the technology works exactly as well as security teams feared. The question this indictment actually answers isn’t “can AI fake a CEO’s voice convincingly.” It’s “what, specifically, stops the fraud once it can” — and the indictment hands over the answer almost by accident.

On November 6, prosecutors charged three men in a seven-count indictment for using AI to clone the voices and faces of real executives to trick accounting staff into wiring money to accounts scattered across nine countries. CBS News, NBC News, and PBS NewsHour all confirmed the indictment independently the week it broke. The scheme ran for roughly 18 months, targeted thousands of companies, and cleared nine figures before anyone was arrested.

Quick Summary: What Happened

Detail	Info
Charged	Jacob Soall, 41; Sharon Hazan, 54; Guy Sheetrit, 42
Where filed	U.S. District Court, Southern District of New York
Charges	Conspiracy to commit wire fraud, wire fraud, two counts of aggravated identity theft, money laundering conspiracy, two counts of money laundering
Maximum penalties	Up to 20 years each on the wire fraud and money laundering counts; aggravated identity theft carries a mandatory 2 years
Timeline	Scheme ran from roughly July 2024 to February 2026
Amount stolen	Approximately $129 million
Victims named	Hospital systems in Missouri and California, a bank in Rhode Island, a UK fintech, a Massachusetts mutual fund company, a Florida pet food company
Official source	U.S. Attorney’s Office, SDNY

Bottom line: The AI voice cloning is the part that makes headlines; the missing callback step and the missing second sign-off are the part that actually let $129 million walk out the door.

What Actually Happened

The indictment describes a scheme that’s almost old-fashioned in its setup and brand new in its execution. Business email compromise — someone posing as a CEO, pushing an urgent wire transfer — has been a staple of corporate fraud for over a decade. What changed is the second step. Instead of relying on a convincing email alone, prosecutors say Soall, Hazan, and Sheetrit followed up with a phone call or video conference where AI had cloned the real executive’s voice, and sometimes their face, well enough to pass as the genuine article to an accounting employee under pressure.

That’s the detail that separates this from garden-variety BEC fraud. An email can be scrutinized, forwarded to a colleague, held for a day. A phone call from someone who sounds exactly like your CEO, asking you to help close a confidential deal right now, doesn’t give an accounting clerk that same luxury of pause.

Per the SDNY’s own press release, the government’s language for the underlying technology is “super intelligence,” or “SI” — which is either prosecutorial flourish or borrowed straight from how the defendants described their own tooling. Either way, strip the branding and it’s the same thing security researchers have been flagging for two years: commodity voice and video synthesis, good enough to defeat the human instinct that recognizing a familiar voice means something.

A Timeline of One Attack

July 2024 — the first targets. Prosecutors say the scheme began running against companies roughly this far back, well over a year before the first arrest.

The inbox. An accounting employee at a targeted company received an email that appeared to come directly from their CEO, flagging a confidential financial matter that needed quiet, fast handling.

The call. A phone or video call followed. The defendants allegedly used AI to reproduce the real executive’s voice — and in some cases their on-screen likeness — closely enough that the employee believed they were talking to their actual boss.

The lawyer. The “CEO” looped in outside counsel to vouch for the transfer. Per the indictment, that attorney was also a defendant, impersonating a real lawyer at a major international firm and allegedly carrying a handwritten script titled “Lawyer Call.”

The wire. Convinced by two separate authority figures in the same afternoon, the employee authorized the transfer to an account the conspirators controlled.

The trail. Prosecutors say the money then moved through accounts in the U.S., Spain, Portugal, Greece, the UAE, Singapore, Austria, Slovakia, and China before conversion into cryptocurrency — a laundering chain built to outrun anyone trying to trace it.

June–September 2026 — the arrests. Soall was arrested in Spain, Hazan in Spain, and Sheetrit in the UK, roughly four months after the scheme is alleged to have stopped running.

What Are Your Options Now

If you touch wire approvals, vendor payments, or executive-requested transfers at your company, this case is a reason to actually check your controls rather than filing it under “scary AI news.”

• Add a callback requirement that doesn’t rely on voice recognition alone. If a “CEO” calls or video-conferences requesting an urgent wire, the verification step needs to be a callback to a known number, or an in-person or pre-established secure-channel confirmation — not “it sounded like them.”
• Build in friction for urgency, on purpose. Every account in this case followed the same pattern: confidential, time-sensitive, don’t tell anyone else yet. That combination should be a hard stop in your approval process, not a reason to skip steps.
• Treat “I’ll connect you with our attorney” as a red flag, not reassurance. The scheme specifically used a second impersonated authority figure to make the first one credible. A legitimate urgent request doesn’t typically route through a law firm your team has never worked with before.
• Know who actually has standing authority to approve a wire. If a single employee can move six or seven figures off a phone call with no second sign-off, that’s the actual vulnerability here, independent of how convincing the AI gets.
• Read up on where AI liability and insurance coverage currently stand. Coverage for AI-enabled fraud losses is still a developing area, and the insurance gap for AI-related liability is wider than most finance teams assume.

Why This Matters

Thousands of targeted companies, over 18 months, for a combined $129 million. That’s not a handful of whales — it’s a volume operation built to clear often enough across a lot of targets rather than needing one perfect hit.

The victim list makes the point better than any warning could. This didn’t hit Silicon Valley unicorns with security budgets to match. It hit a bank in Rhode Island, hospital systems in Missouri and California, a pet food company in Florida, a mutual fund manager in Massachusetts, a UK fintech — ordinary mid-market organizations with a controller and a couple of accounting staff, not a dedicated fraud team. That’s the real audience for this story: not Fortune 500 CFOs with deepfake-detection vendors on retainer, but the small and mid-size finance teams this site’s audience actually represents.

Sharon Hazan’s alleged role is the detail that’s easy to skim past. Per NBC News, Hazan handled the laundering side and bragged of cleaning more than $400 million over three years — over three times the $129 million tied to this specific indictment. That’s not three guys who stumbled onto a clever trick. That’s a laundering operation sized to absorb proceeds at scale, and this case is one slice of a bigger business.

We covered banks warning about AI shopping agents opening new scam vectors back in September. The underlying problem here is the same one wearing a different costume: a wire-approval process built around “does this email look legitimate” was never designed for “does this voice on the phone sound exactly like the person it’s supposed to be.”

Our Take

Run the numbers on what would have actually stopped this, and the AI barely factors in. A $250,000-a-month regional distributor with a standard controls setup — single-employee sign-off up to $10,000, two signers above that, and no callback requirement for “urgent” requests — is exactly the profile this scheme hunted for. Add one rule — any wire framed as confidential or time-sensitive triggers a callback to a number already on file, no exceptions — and a flawlessly cloned voice asking for a six-figure “confidential acquisition deposit” has nowhere to go. The call still happens. The money doesn’t move. That’s the headline argument here, not a footnote: process controls stopped this fraud pattern decades before AI existed, and they’re still what stops it now. Voice-detection software is chasing a synthesis-quality arms race it will never definitively win; a callback and a second signature don’t need to win an arms race, because they don’t care whose voice is on the line.

That reframes the rest of the case. Soall allegedly carried a literal script for impersonating a UK attorney — a second impersonated authority figure, deployed specifically to make the first one credible. Legal AI and law firm workflows that verify identity on their end would blunt that angle too, for the same reason a callback blunts the CEO call: it doesn’t depend on recognizing a voice, it depends on a process step nobody can talk their way around.

None of this required exotic technology, either. The voice and video synthesis behind the scheme is the same underlying capability powering legitimate products that help call centers and creators clone voices with consent and disclosure. We’ve also covered frontier AI models lying to investigators in adversarial testing and a zero-click exploit hitting AI coding agents — both cases where the eventual fix was a structural control, not a smarter detector, sitting in front of the AI rather than trying to out-think it. This case is the same lesson with nine figures attached: the money didn’t move because the AI was convincing. It moved because nothing in the approval chain cared whether it was convincing in the first place.

Frequently Asked Questions

Who was charged in the AI voice-cloning CEO fraud case?

Jacob Soall, 41, Sharon Hazan, 54, and Guy Sheetrit, 42, were charged in a seven-count indictment unsealed in the U.S. District Court for the Southern District of New York on November 6, 2026.

How much money did the scheme steal?

Prosecutors say the scheme stole approximately $129 million from victim companies over roughly 18 months, between July 2024 and February 2026.

What charges do the defendants face?

Conspiracy to commit wire fraud, wire fraud, two counts of aggravated identity theft, money laundering conspiracy, and two counts of money laundering. The wire fraud and money laundering counts each carry a maximum of 20 years; the aggravated identity theft counts each carry a mandatory 2 years.

How did they use AI to impersonate CEOs?

According to the indictment, the defendants used AI to clone the voice, and sometimes the video image, of a real company executive on phone and video calls with accounting employees, directing them to wire funds for a supposedly confidential transaction.

Who were the victims?

Named victims include hospital systems in Missouri and California, a bank in Rhode Island, a financial technology company in the United Kingdom, a mutual fund company in Massachusetts, and a pet food company in Florida.

Were the defendants arrested?

Yes. Soall was arrested in Spain on June 17, 2026; Hazan was arrested in Spain on August 5, 2026; and Sheetrit was arrested in the United Kingdom on September 7, 2026. Extradition requests are pending for all three.

Is this the first prosecuted case of AI voice-cloning fraud?

It’s the first case to attach a figure this large — $129 million — to an AI-powered executive impersonation scheme prosecuted by U.S. federal authorities, confirming the risk is an active, organized criminal practice rather than a theoretical one.

How can my company protect itself from this type of fraud?

Require callback verification through a known, pre-established channel for any urgent wire request, mandate second sign-off above a set dollar threshold, and treat requests for confidentiality or urgency as a reason to slow down rather than speed up — regardless of how convincing the voice or video on the call sounds.

---

Last updated: November 9, 2026. Sources: U.S. Attorney’s Office, Southern District of New York · CBS News · NBC News · PBS NewsHour · SecurityWeek.

Related reading: Major Banks Warn AI Shopping Bots Enable New Scams · AI Insurance Gap: Labs Face Billions, Coverage Caps at $50M · AI Tools for Lawyers: The 2026 Reality Check · Best AI Tools for Small Business · ElevenLabs Hits $500M ARR on Institutional Voice AI