Three charged in the US with stealing $129 million using AI clones of executives' voices and faces

Imagine an ordinary workday for an accountant. A director whose face and voice they know appears in a messenger app or on a video call. The company's lawyer is alongside the director. They say the deal is urgent and confidential and ask the accountant to make a transfer. Everything looks and sounds legitimate. According to US prosecutors, this is precisely the scenario used by the three defendants. The result: about $129 million.

What is known about the case

The indictment was unsealed on Thursday, the US Attorney's Office in Manhattan said. The defendants are Jacob Soall, 41, Sharon Hazan, 54, and Guy Shitrit, 42. Investigators allege that they used AI to clone the voices and faces of company executives and lawyers, deceiving employees who then transferred the money. Soall was arrested by Spanish police.

An important caveat: these are charges, not a conviction. The published excerpt does not disclose exactly which tools the alleged perpetrators used, how many companies were affected, or how the money was moved. Those details will emerge in the case materials, and it is too early to draw conclusions about the techniques involved.

Why this is not ordinary phishing

A conventional “urgent payment” attack relies on text: a forged email, an unfamiliar address, or awkward wording. Employees have long been trained to be wary of such attempts. A voice and a face command a different level of trust. We are accustomed to treating a video call with a manager as proof of identity. Voice and facial cloning breaks that assumption.

The second problem is the combination of roles. When a “director” and a “lawyer” both take part in a conversation, employees may assume that the process has already been approved and that their only job is to carry it out. The pressure of authority and urgency can overpower any security guidance if the company lacks a strict procedure.

The uncomfortable conclusion for the industry is that “I saw and heard the person myself” can no longer be treated as proof of authenticity. This applies not only to bank transfers but also to access resets, key handovers, deal approvals, and any action authorized on someone's word.

What teams should do

The following are general defensive practices, not details of this case.

• Verify through a second channel. Confirm every transfer or change to payment details through an independent route: call back using a number from the internal directory, not the number from which the request came.
• Use the two-person rule. Large payments should require approval from two people, neither of whom should give in to pressure to act “right now.”
• Use code words and verification questions. Establish verification methods in advance for senior executives and finance teams that cannot be found in public sources.
• Set limits and delays. Require a mandatory pause for unusual recipients and large amounts. Fraud thrives on haste, while a delay creates time to spot the trap.
• Train with live examples. Let employees hear what a cloned voice sounds like. One such demonstration is more persuasive than ten slides.

Companies building fintech products, HR platforms, video communications systems, and onboarding tools should account for this in their architecture. Identity verification through video or voice alone can no longer be considered reliable. Additional signals are needed, including the device, action history, and in-app confirmation.

Limitations and conclusion

The source text is brief, so we do not know how convincing the fakes were, whether they operated in real time, or how much money was recovered. A single case does not mean that every video call is dangerous. But the $129 million figure shows that AI impersonation attacks are no longer merely a laboratory threat: they can cause extremely large losses.

The practical conclusion is simple: if your process rests on the phrase “I recognized their voice,” you have a vulnerability. Close it with a procedure before someone else discovers it.