Federal prosecutors in Manhattan say three men used artificial intelligence to sound and look like chief executives and big-firm lawyers, then talked accounting staff into wiring money to accounts the conspirators controlled. An indictment unsealed in the Southern District of New York alleges the scheme ran from July 2024 through February 2026 and took more than $129 million from companies in the United States and abroad. The defendants are presumed innocent. Every dollar figure, victim description and boast quoted below comes from charging documents and official statements, not from proven facts at trial.
TL;DR
- Prosecutors charged Jacob Soall, Sharon Hazan and Guy Sheetrit in SDNY with an alleged AI-assisted CEO and attorney impersonation scheme said to have stolen more than $129 million between July 2024 and February 2026.
- The indictment describes deepfake voice and image calls, fake “confidential” deals, impostor lawyers directing wires, then crypto conversion across accounts in multiple countries.
- Arrests were made in Spain and the United Kingdom; extradition requests are pending, and the charges remain allegations until proven in court.
Who was charged, and with what
According to reporting based on the U.S. Attorney’s Office for the Southern District of New York, the defendants are Jacob Soall, also known in the case as “Rocky Balboa” and “Seba Clarke”; Sharon Hazan, referred to with nicknames including “Superman Cash,” “Impossible Mission,” “Fat Boy” and “Ron”; and Guy Sheetrit. Soall, 41, of Spain and the United Kingdom, was arrested in Spain on June 17, 2026. Hazan, 54, of Israel, was arrested in Spain on August 5, 2026. Sheetrit, 42, of Israel and the United Arab Emirates, was arrested in the United Kingdom on September 7, 2026. Extradition requests for all three are pending.
Each faces seven counts, as summarised by prosecutors and press accounts of the indictment: conspiracy to commit wire fraud, wire fraud, two counts of aggravated identity theft, money laundering conspiracy, and two counts of money laundering. Maximum exposure is serious; press summaries differ on how the statutory maxima stack, so treat any “104 years” headline as a stacking of maxima rather than a predicted sentence. Aggravated identity theft carries its own mandatory term under federal practice. None of that is a conviction.
U.S. Attorney Jamie McDonald said the defendants allegedly “abused groundbreaking super intelligence technology to impersonate real people and deceive corporate executives into sending them and their co-conspirators approximately $129 million,” using the White House’s preferred label for advanced AI. FBI Assistant Director in Charge James C. Barnacle Jr. of the New York Field Office called the alleged scheme an example of how advanced technology can be exploited for large-scale fraud and said the Bureau is adapting to disrupt those who harm New Yorkers and businesses in the district.
How the alleged playbook worked
Prosecutors describe a multi-step social-engineering path aimed at employees with authority to approve large wires, usually in accounting. Co-conspirators allegedly posed as the company’s CEO and claimed help was needed on a confidential financial transaction, often framed as an acquisition or similar deal. An email from an e-signature service sometimes opened the contact, with the impersonated CEO asking the employee to sign a nondisclosure agreement before speaking with counsel, according to accounts of the unsealed indictment.
On voice or video calls, the group allegedly used AI to mimic the CEO’s voice and, at times, image. The “CEO” then directed the employee to attorneys at major international law firms who were supposedly working the deal. Those attorneys, prosecutors say, were also impostors—often Soall reading from a handwritten script titled “Lawyer Call” while impersonating United Kingdom counsel. The fake lawyer then instructed the employee to wire a large sum to a bank account that, in reality, belonged to the conspiracy.
After the wire landed, prosecutors allege, the money moved through accounts in places including the United States, Spain, Portugal, Greece, the United Arab Emirates, Singapore, Austria, Slovakia and China, then into cryptocurrency to obscure source, ownership and control before distribution among co-conspirators. Hazan is accused of leading laundering and of boasting that he had laundered money since 1994, more than $400 million in the past three years, and was “number one all over the world” among professional launderers—allegations drawn from the government’s case, not admissions in open court. Prosecutors also note Hazan’s 2008 Southern District of New York conviction for conspiracy to commit bank fraud. Sheetrit is likewise accused of laundering proceeds.
Victim companies, named only as the indictment describes them
Public charging summaries list industry and location descriptors and loss figures; they do not require journalists to invent trade names. Alleged losses highlighted in coverage include roughly $11.58 million from a United Kingdom financial-technology company, $5.24 million from a Florida pet-food company, about $4.83 million from a Missouri hospital system, about $4.64 million from a Massachusetts mutual-fund company, nearly $3.0 million from a California hospital system, about $2.8 million from a New York investment bank or investment company, and about $2.68 million from a Rhode Island bank. Those figures are government allegations. Company identities beyond those descriptors were not consistently published in the primary press accounts reviewed for this piece, so they are not guessed here.
Prosecutors also say the conspiracy targeted employees across many organisations and impersonated attorneys at some of the world’s largest law firms. “Thousands of victims” appears in some case summaries as a government characterisation of the targeting effort; that is not the same as thousands of completed thefts. The $129 million figure is the aggregate loss the indictment attributes to successful wires.
Why this lands in the privacy and compliance lane
Deepfake-enabled payment fraud is not only a cybercrime story. It sits at the intersection of identity, voice biometrics, vendor payment controls and board-level AI risk. Governance debates about when a synthetic voice creates legal risk, litigation over AI voice calls and consent gaps, and even lawsuits over voice-training pipelines all point to the same operational fact: a familiar voice on a rushed call is no longer reliable authentication.
Insurers and security vendors have been warning about deepfake wire fraud for several years, including products marketed against deepfake-driven financial loss. Platform fights over generative video and deepfake backlash, and regulator actions such as privacy findings tied to deepfake misuse, show how quickly synthetic media moved from novelty to enforcement context. This indictment, if the allegations hold up, is the cash-out version of that trend: not a celebrity deepfake for clicks, but a CFO’s worst afternoon.
Six controls that would have interrupted the alleged script
None of the following depends on waiting for extradition hearings. They are payment-integrity and identity checks many treasury and privacy teams already own on paper and underuse under time pressure.
- Out-of-band CEO confirmation. Any request to move unusual sums, especially under an NDA or “confidential deal” framing, gets a callback to a known phone number or an in-person/chat confirmation on an existing authenticated channel—not a return call to a number supplied in the same thread.
- Dual control that ignores rank on the call. Wire changes and new beneficiary accounts require a second approver who was not on the deepfake call. Executive urgency is a social-engineering feature, not a control override.
- Attorney verification against firm directories. If “outside counsel” suddenly appears on a deal, verify the partner through the law firm’s switchboard and known email domain. A scripted British-lawyer persona is cheap; a real firm’s operator is not part of the plot.
- Beneficiary cool-off for new accounts. New international accounts, especially in jurisdictions outside the company’s normal corridor, sit in a hold queue measured in hours or days, not minutes.
- Deepfake-aware training with real call examples. Staff who can move money need drills that include cloned-voice audio, not only phishing emails. Record that training; after a loss, “we mentioned AI once in a slide” will not help.
- Incident playbooks that preserve the call. If something feels wrong mid-transfer, freeze the wire, preserve the e-signature trail, cloud-meeting recording and SIP logs, and escalate to fraud and privacy leads together. Crypto conversion starts quickly once funds clear.
Reporting from NBC News and CyberScoop tracks the indictment’s narrative closely; organisations that may have been contacted in similar patterns are being urged by the FBI to come forward.
Frequently asked questions
Have the defendants been convicted?
No. They have been charged. Arrests abroad and pending extradition do not equal guilt. The presumption of innocence applies.
Did prosecutors name every victim company?
Public summaries emphasise industry, location and loss amounts. This article does not invent trade names that those summaries did not clearly publish.
Is “super intelligence” a different technology from AI?
In the government’s statements it is branding for advanced artificial intelligence. The alleged tools are described as AI used to mimic voices and images.
What should a finance team change this week?
Require out-of-band verification for any executive-directed wire, especially new beneficiaries and cross-border accounts, and test that rule with a simulated cloned-voice call.
The call that should never be enough
For a decade, security awareness training treated the urgent CEO email as the classic trap. The SDNY case, as alleged, updates the trap for an era when the voice on the line can be synthetic and the lawyer on the follow-up can be the same crew with a better accent. If prosecutors prove even part of the $129 million narrative, the expensive lesson will not be that AI is powerful. It will be that companies still treat a familiar voice plus a confidential-sounding story as sufficient authority to move eight-figure sums. That habit is now a balance-sheet risk.
