
Blogorganization
When the Boss Calling Isn't Real: $129M in AI Voice Fraud
An accountant at a Missouri hospital system picked up the phone, heard the voice of her organization's chief executive, and within hours authorized a $4.8 million wire transfer. Both the CEO and the lawyer who followed up were AI-generated voices, according to the U.S. Attorney's Office for the Southern District of New York.
The Missouri transfer was not a one-off. Prosecutors say the same approach ran repeatedly over roughly 18 months beginning in July 2024, hitting companies in the United States and Europe for more than $129 million in total. A British financial technology company lost $11.6 million. A Florida pet food company lost $5 million. A Massachusetts mutual fund company lost $4.7 million. A California hospital system lost $3 million. A Rhode Island bank lost $2.7 million. Three men arrested in Britain and Spain are awaiting extradition on charges of wire fraud, aggravated identity theft, and money laundering, according to an indictment unsealed November 5. The charges are allegations.
The mechanics were consistent. A finance manager or accountant received an email borrowing the name of a real attorney at a large law firm. A phone call followed, with an AI-cloned executive voice asking for an urgent, confidential transfer. A second call, posing as outside counsel, applied legal pressure. A handwritten script, later recovered by investigators, leaned on regulatory language about insider trading to justify secrecy, specifically to prevent staff from checking with colleagues. The money moved to accounts across multiple countries and was converted to cryptocurrency.
The secrecy demand was the load-bearing element. Once a finance employee was told not to discuss the request internally, the normal friction of approval chains disappeared. Prosecutors also noted attempts to pull employees into video calls with deepfake versions of executives, suggesting the operation was scaling its impersonation techniques beyond audio.
Why the Voice Alone Is Not Enough
Hany Farid, a computer science professor at Dartmouth College and chief science officer at GetReal Security, told The New York Times: "I would comfortably say with less than a minute, you can clone somebody." On whether employees can tell the difference, he added: "People are just slightly above chance performance."
That framing matters for how finance teams think about authorization. If the voice of a CEO cannot be trusted on a call, the call itself cannot be the authorization. It needs to be one input into a process that includes a check the caller must pass before any transfer moves.
Trust Onion's rotating codewords give finance teams exactly that check. Before approving a wire transfer or vendor payment change, a finance team member asks: "What are the words?" The real executive knows the current three codewords, which rotate every 60 seconds. An AI voice clone cannot know words it never heard.
The FBI's Internet Crime Complaint Center's 2025 report logged more than 22,000 complaints related to artificial intelligence, totaling over $890 million in losses across many kinds of schemes, including phishing emails, voice-clone scams, and romance fraud. That figure covers a broad range of AI-related fraud, not voice cloning alone.
What Would Have Changed the Outcome
The Missouri accountant had no process to challenge the caller before moving money. The secrecy instruction, embedded in the lawyer script, was designed to close off every other path she might have taken. A codeword check runs in seconds, requires no explanation to a legitimate executive, and cannot be blocked by an instruction to stay quiet, because it happens before the transfer is approved, not after.
After the arrests, one chief executive described to Farid what his organization planned to change about executive communication. "That's it," he said. "We're doing everything in person."
In practice, that is not a policy most finance teams can hold. Executives travel, calls happen across time zones, and urgent decisions do not wait for the next all-hands. The question is not whether to trust the phone. It is whether the caller can pass a check the phone cannot fake.
A controlled voice-cloning safety demonstration
Try our tool to see just how easy voice cloning is. This controlled safety demonstration exists only to show how easy voice cloning is and to make the risk tangible. It is not a real incident or proof of anyone's identity. Your submitted sample is handled transiently to generate the fixed demonstration audio. Trust Onion does not identify you or the person from the sample, and does not keep the submitted recording or generated audio. The temporary voice clone is deleted from our voice provider as soon as the demo finishes; if that deletion fails, it is retried automatically until the provider confirms it.
Add Trust Onion's rotating codeword check to executive and wire-transfer approvals. Three words, free for your team, at trustonion.io/enterprise.
Protect Your Business FreeCite this page
When the Boss Calling Isn't Real: $129M in AI Voice Fraud — Trust Onion, November 18, 2026. https://trustonion.io/blog/ai-ceo-voice-clone-wire-fraud-129-million



