CEO and lawyer deepfakes: $129M stolen, as voice and image fraud now operates at industrial scale

By The Guardian (with additional reporting from The Jerusalem Post) · November 6, 2026.

Three men, Jacob Soall, Sharon Hazan and Guy Sheetrit, have been arrested and charged in the US with using AI to impersonate the voices and likenesses of executives and lawyers to get employees to transfer large sums. According to the indictment unsealed by the Department of Justice (DoJ) through the US Attorney's Office for the Southern District of New York, the haul exceeds $129 million. These are allegations, not convictions: the defendants are presumed innocent.

According to the DoJ, the scheme worked in two steps. First, the defendants posed as a senior executive and asked an employee for help with a sensitive financial transaction. They then referred the employee to supposed lawyers at a major international law firm, also impersonated, who gave instructions on where to send the money. The double layer of authority, executive plus lawyer, lent credibility to a request that would otherwise have raised suspicion.

The facts, as the source reports them: the alleged fraud ran from 2024 to 2026. Victims include a Missouri hospital system (nearly $5M), a Rhode Island bank (about $2.7M), a California hospital system (about $3M), a Massachusetts investment fund (more than $4.6M), a Florida pet food company (about $5M) and a British tech company (more than $11M). The funds were sent to accounts in the US, Spain, Portugal, Greece, the United Arab Emirates, Singapore, Austria, Slovakia and China, and then converted into cryptocurrency to conceal their origin, according to the DoJ.

Soall and Hazan were arrested in Spain this summer and Sheetrit in the UK in September. They are charged with conspiracy to commit wire fraud, money laundering conspiracy and aggravated identity theft.

**Our take.** This case dismantles a comfortable idea: that the danger of AI lies in some distant superintelligence. The documented harm today is far more mundane and far more measurable. Generative AI (models that produce convincing voice, images or text) has driven the cost of faking authority down to almost zero, and authority is the mechanism companies use to approve payments.

The figure matters less for its total than for its spread: hospitals, a bank, a fund, a pet food company. None of these organizations is an obvious target for state espionage; they are simply businesses with cash and processes that rely on recognizing a voice or a face. When that signal stops being reliable, the control fails everywhere at once.

In general, the defense will not come from asking employees to “spot” deepfakes by ear or by eye, a race people tend to lose as the technology improves. It will come from redesigning processes: verification through an independent, pre-agreed channel, dual approval for transfers, limits and delays on new or international payments, and default distrust of urgency and secrecy. This is process governance, not model governance.

There is also a regulatory lesson. Banning voice synthesis technology is neither feasible nor desirable: it is the same foundation that gives a voice back to those who have lost theirs and makes dubbing and accessibility cheaper. What works, as this case shows, is cross-border police cooperation (arrests in Spain and the UK over a New York case) and financial traceability. Following the money works better than chasing the algorithm.

In the short term, the outlook is uncomfortable: we expect more cases like this and a painful transition until companies adapt their controls. In the long term, the same AI that serves fraud today will make it possible to verify identities and detect financial anomalies at a scale impossible for humans. The abundance we promise first needs a solid infrastructure of trust, and this case is a reminder that we do not have one yet.