AI Voice Cloning Fraud: Inside the $129M CEO Impersonation
$129 Million, Two Cloned Voices, One Missing Callback
Federal prosecutors in Manhattan have charged three people with using AI voice cloning to impersonate CEOs and big-firm lawyers, allegedly steering employees at hospitals, a bank, a fintech and a mutual fund company into wiring more than $129 million. Here is how the scheme worked, why the fake lawyer was the masterstroke, and the verification procedure that stops it.
The fraudsters did not only fake the CEO. They faked the verification, too.
What prosecutors allege in the AI executive impersonation case
On November 5, 2026, federal prosecutors in Manhattan announced charges against three defendants accused of running an AI voice cloning fraud operation from July 2024 through February 2026. According to the charges, the group used artificial intelligence to clone the voices of corporate executives and of attorneys at large international law firms, and on video calls used AI to mimic a CEO’s voice or image.
Prosecutors say losses exceeded $129 million across numerous companies. The defendants were arrested in Spain and the United Kingdom between June and September 2026 and are awaiting extradition. They face conspiracy, money laundering and related charges. These are allegations, and every defendant is presumed innocent unless proven guilty in court.
in losses alleged by prosecutors, across numerous companies
cloned personas per scheme: the CEO and the outside lawyer
months the scheme allegedly ran, July 2024 to February 2026
The named victims read like a cross-section of the economy, and several of them are heavily regulated organizations with mature security programs.
| Victim named in the charges | Reported loss |
|---|---|
| U.K.-headquartered fintech company | $11,578,474 |
| Florida pet food company | $5,242,747 |
| Missouri hospital system | $4,829,580 |
| Massachusetts mutual fund company | $4,639,918 |
| California hospital system | $2,968,497 |
| Rhode Island bank | $2,684,580 |
Six of the victims named in public reporting. Prosecutors say the indictment cites others. Source: ABC News.
How the AI voice cloning fraud scheme worked
The playbook, as prosecutors describe it, is a classic executive impersonation chain with one important upgrade: a second trusted persona whose only job was to make the first one believable.
Every confirmation in the chain came from a source the fraudsters controlled.
The cloned CEO makes contact
An employee hears their own chief executive. The caller asks for help with a confidential financial transaction. Authority, urgency and secrecy arrive together in the first minute.
Secrecy closes the normal path
“Confidential” does real work here. It gives the employee a reason not to loop in their manager, the treasury team or anyone else who might ask the obvious question.
A cloned lawyer confirms the deal
The CEO directs the employee to outside counsel at a large international firm. The employee reaches out, hears a second professional voice, and receives what feels like independent confirmation.
The wire goes out
Large sums move to accounts around the world. At no point did anyone check the request against a phone number, contact or approval path that the requester had not handed over themselves.
Why the fake lawyer was the masterstroke
Most finance teams are taught a simple rule: if something feels off, confirm it with someone else. This scheme turned that instinct into the delivery mechanism. The employee did seek a second opinion. The second opinion was simply supplied by the same people running the fraud.
That is the difference between corroboration and verification, and it is the single most important lesson in this case.
The CEO sounds right. The lawyer sounds right. The two stories match. Every one of those checks relies on contacts and context the requester chose to provide, so a well-built pretext passes all of them at once.
A callback to the CEO on the number in the corporate directory, and to the law firm through the main line on its own official site. It uses information the requester never touched, so it gives the same answer whether the cloned voice is crude or flawless.
Why spotting the fake would not have saved these companies
It is tempting to read this case as an argument for teaching employees to hear the tells in a synthetic voice. That is exactly why we argue that deepfake detection training is a decaying control. Human perception is a moving target. Every model release makes the voice cleaner, the latency lower and the video steadier, and the employee is being asked to win that race on a live call under pressure from their own boss.
Technical identity verification is a different story, and a valuable one. Strong authentication on payment platforms, verified callback directories and dual-control approvals do not get weaker as voice synthesis improves. A procedure tied to the request type, not to how the caller sounds, is invariant to generation quality. That is the control worth building, and the control worth testing.
A workforce that invokes process defeats our red teams. Users who only spot fakes do not, consistently.
The controls that stop AI voice cloning wire fraud
- Call back on a number you already hold. Reach the executive through the corporate directory and outside counsel through the firm’s official main line, never through a number, email or meeting link supplied in the request.
- Confidentiality never waives verification. Write it into policy: a request to keep a transaction secret from normal approvers is itself a trigger for escalation.
- Bind verification to the request type. Every wire, new beneficiary, payment-detail change and urgent transfer gets the same check, however senior the caller sounds.
- Require dual control above a set threshold. The second approver verifies independently instead of trusting the first.
- Hold first payments to new accounts. A beneficiary the organization has never paid before, especially offshore, waits for verification.
- Treat video the same as voice. A face on a call is a channel, not a credential. The callback rule applies to Teams and Zoom requests too.
- Make pausing blameless. Real executives and real lawyers expect to be called back. Nobody is penalized for delaying a genuine deal to verify it.
What to measure
A phishing click rate would never have surfaced this risk. The questions that matter are about whether the procedure exists and whether it runs.
- Coverage. Does every funds-movement request type have a written, channel-independent verification requirement?
- Process hold rate. When a senior voice asks for a confidential wire, does the callback actually happen?
- Exception rate. How often is verification skipped because the request came from the top, or because a second voice confirmed it?
- Time to verify. How long the callback takes under deadline pressure, because a slow procedure is the one that gets skipped.
How Breacher.ai tests executive impersonation risk
This is exactly the chain Breacher.ai was built to rehearse. Our OSES™ engine runs authorized, orchestrated social engineering simulations that combine live, two-way AI voice, deepfake video and Teams or Zoom meetings with email and SMS, adapting each step to how your team responds, all the way to the consequential action. That includes the hand-off from one trusted persona to another, the part of this case that made it work.
From finding to fix, on one platform
Findings are named against procedures, not people. You see which callbacks ran, which request types had no verification step at all, and where a confidentiality pretext pulled someone off the normal path. OSES™ Behave then generates training on the exact procedure that slipped, aligned to your own policies, in minutes. The next simulation proves whether the control now holds.
Executive and payment scenarios are the core of our CEO fraud simulation. Voice-led chains run through our vishing simulation and video-led chains through our deepfake simulation, all on the Breacher.ai simulation platform. For what these calls look like in the field, read our AI vishing red team field findings.
Measure your risk.
Train for what you find.
Prove it changed.
Find out whether a cloned CEO and a fake lawyer would get a wire past your finance team, before a real adversary tries.
Book Your DemoFrequently asked questions
AI voice cloning fraud is a form of social engineering in which criminals use artificial intelligence to reproduce the voice of a trusted person, such as a CEO, a lawyer or a vendor, and then use that voice on a call to persuade an employee to wire money, change payment details or share access.
According to federal prosecutors in Manhattan, the group used cloned voices of chief executives to ask employees for help with a confidential transaction, then directed them to lawyers at large international law firms. Those lawyers were also impersonated with cloned voices, and the employees were told to wire large sums to accounts the group controlled.
The fake lawyer gave the employee what felt like independent confirmation. In reality, the second voice was reached through details supplied by the first, so the confirmation came from the same people running the fraud. That is corroboration, not verification.
Not reliably, and less so with every model release. Human perception of synthetic voice is a decaying control. Technical identity verification and procedural controls, such as a callback to a number already on file, hold regardless of how convincing the voice sounds.
A callback to the executive on the number in the corporate directory, and to any outside counsel through the firm’s official main line, made before money moves and never to a contact supplied in the request. Pair it with dual approval above a threshold, a hold on first payments to new accounts, and a rule that confidentiality never waives verification.
Breacher.ai runs authorized, orchestrated social engineering simulations with its OSES™ engine, combining live two-way AI voice, deepfake video, Teams and Zoom meetings, email and SMS. It carries executive impersonation chains to the consequential action, names findings against procedures rather than people, and generates targeted training on the procedure that slipped.
Test the Callback Before Someone Else Does
Thirty minutes. We show you how a cloned executive and a second trusted voice would reach your finance team, and the verification step that would stop the wire.
Book Your DemoLatest Posts
Table Of Contents
About the Author: Jason Thatcher
support@breacher.ai
