Every voice deepfake attack that mattered, 2019 to 2026
Fourteen confirmed cases of cloned voices used to steal money, sway an election, ruin a reputation or impersonate a government, in date order, each with a primary source.
Short answer
Criminals have used cloned voices to steal money since at least 2019. The largest confirmed loss is about 95 million euros, taken from the Italian bank Fideuram in February 2026 after a call in the cloned voice of a real lawyer. Other confirmed cases include a 25.6 million dollar transfer at Arup in 2024, a 35 million dollar transfer in Hong Kong in 2020, and thousands of smaller frauds against families and bank customers. Cloned voices have also turned up in Slovakia's 2023 election, in a robocall telling New Hampshire voters to stay home, in a faked rant that cost a school principal his job, and in voicemails impersonating the US Secretary of State. This page lists each case that has a court record, a police statement or a primary news report, in date order. We add cases as they are confirmed.
What is a voice deepfake attack?
A voice deepfake attack is any fraud, manipulation or impersonation that relies on synthetic speech made to sound like a specific real person. The attacker collects a few seconds to a few minutes of the target's voice from interviews, earnings calls, social media videos or voicemail, feeds it to a voice cloning model, and then either plays prepared recordings or speaks live through the clone. When the call is used to extract money or credentials, security teams call it vishing, short for voice phishing.
A familiar voice changes how you treat the call. If your finance director asks you to wire money, you skip the checks you would run on a stranger. The attacker also pays almost nothing for that trust: a magician made the 2024 Biden robocall in under twenty minutes for 150 dollars.
The cases, in order
Each entry gives the date, the target, what the cloned voice was used for, the outcome and the source. Cases marked "not confirmed" appear because other reports cite them so often, but no authority has verified that AI was used.
UK energy company, about 220,000 euros
- TargetChief executive of a UK energy firm
- Loss220,000 euros, not recovered
- ChannelPhone call
The chief executive of a British energy firm received a call in the voice of his boss at the German parent company, accent and cadence included. The caller told him to wire 220,000 euros to a Hungarian supplier within the hour, and he did. The insurer Euler Hermes said the attackers used commercial voice generation software. The money moved through Hungary and Mexico and the firm did not get it back. This is the first voice clone fraud on record.
Source: Wall Street Journal, August 2019
Hong Kong branch of a Japanese company, up to 35 million dollars
- TargetBranch manager
- LossUp to 35 million dollars
- ChannelPhone call plus emails
A branch manager in Hong Kong took a call from a man whose voice he recognised as a director of the parent company. The director said an acquisition was closing and transfers needed to go out. Emails from the director and from a lawyer handling the deal were already in the manager's inbox. He authorised transfers totalling around 35 million dollars. The case became public through a court filing in which the United Arab Emirates asked US investigators to help trace 400,000 dollars that landed in American accounts. Many articles describe this as a bank being robbed; the filing describes the victim as a company and the investigation as UAE-led.
Source: Forbes, October 2021
Arizona, fake kidnapping
- TargetA parent
- LossNone
- StatusUse of AI not confirmed
- ChannelPhone call
Jennifer DeStefano picked up a call and heard her 15 year old daughter sobbing, then a man demanding one million dollars. Her daughter was on a ski trip and safe. Another parent called the police while the call was still running and no money was sent. DeStefano later testified to the US Senate. Police never confirmed that AI was used and never identified a suspect. It remains the most cited example of the "distress" scam, which the FBI now counts as its own category.
Source: CNN, April 2023
Retool, 27 customer accounts taken over
- TargetRetool employee
- Outcome27 customer accounts breached
- ChannelText message, then phone call
Employees of the software company Retool received text messages pretending to come from the IT team during a planned login migration. One employee clicked and entered credentials on a fake portal. Then someone phoned them, speaking in the deepfaked voice of a real colleague and familiar with the office layout and internal processes. The employee handed over one additional multi-factor code. That was enough to take over the account and, through it, the accounts of 27 cloud customers, all in cryptocurrency. Retool published its own account of the attack.
Source: Retool engineering blog, September 2023 and TechTarget
Slovakia, election audio
- TargetProgressive Slovakia and its leader
- OutcomeSpread unchecked during the media silence
- ChannelFacebook and Telegram
Two days before Slovakia's parliamentary election, a 100 second audio clip appeared on Facebook and Telegram. In it, the leader of Progressive Slovakia, Michal Simecka, and a journalist appeared to discuss buying votes from the Roma community. Both denied the conversation and the fact-checking unit of AFP found signs of AI manipulation. Slovakia has a 48 hour media silence before voting, so the journalists and politicians who could have rebutted the clip were barred from speaking. Meta's manipulated media policy at the time covered only video, so Meta left the clip up. Progressive Slovakia lost by a small margin. It is the first documented audio deepfake aimed at a national election in Europe.
Source: AI Incident Database, incident 573 and CEDMO
New Hampshire, the Biden robocall
- TargetDemocratic voters before the primary
- OutcomeThousands of calls, 6 million dollar FCC fine
- ChannelRobocall with spoofed caller ID
Two days before the state's presidential primary, thousands of voters got a robocall in the voice of President Biden telling them to stay home. Political consultant Steve Kramer paid a magician 150 dollars to make the audio with ElevenLabs. It took under twenty minutes. The Federal Communications Commission fined Kramer six million dollars and declared AI voices in robocalls illegal under existing law. ElevenLabs closed the account that made the file.
Source: FCC, September 2024 and NPR
Baltimore, a school principal's voice
- TargetPrincipal of Pikesville High School
- OutcomePrincipal lost his job, maker jailed
- ChannelRecording shared online
A recording spread online in which Eric Eiswert, principal of Pikesville High School, appeared to make racist remarks about students and staff. The school was flooded with calls, Eiswert was suspended, and police patrolled his home after threats. Investigators found the school's athletic director, Dazhon Darien, had made the audio with AI after Eiswert told him his contract would not be renewed. Three forensic experts confirmed the recording was synthetic with added background noise. Darien was sentenced to four months in jail in April 2025. Eiswert left the school.
Source: The Daily Record, April 2025
Arup, Hong Kong, 25.6 million dollars
- TargetFinance employee
- Loss200 million Hong Kong dollars, about 25.6 million US dollars
- ChannelEmail, then video call
A finance employee at the engineering firm Arup received an email from the company's UK chief financial officer about a confidential transaction. He was suspicious until the attackers invited him to a video call with the CFO and several colleagues. Everyone on the call was a deepfake built from public video and audio of the real executives. Reassured, he made 15 transfers totalling 200 million Hong Kong dollars to five accounts. Arup confirmed it was the victim in May 2024 and has not reported getting any of the money back.
Source: CNN, May 2024
LastPass, WPP and Ferrari, three attempts that failed
- TargetEmployees of three large companies
- OutcomeAll three stopped by staff
- ChannelWhatsApp and Microsoft Teams
Three large companies reported attacks in the same season, all with the same shape and none successful.
At LastPass, an employee received WhatsApp calls, texts and a voicemail in the voice of chief executive Karim Toubba. The employee judged that the CEO would not use that channel or that urgency, ignored it, and reported it.
At WPP, attackers set up a Microsoft Teams meeting using a cloned voice of chief executive Mark Read, his public photo, and video pulled from YouTube, and tried to get an agency head to set up a new business and share details. The target noticed inconsistencies.
At Ferrari, an executive got WhatsApp messages and then a live call in the cloned voice of chief executive Benedetto Vigna, southern Italian accent included, about a confidential acquisition. The executive asked what book Vigna had recommended to him a few days earlier. The caller hung up.
Source: LastPass blog, The Guardian on WPP, Bloomberg on Ferrari
Italy, the defence minister's voice
- TargetItalian business leaders
- LossAbout 1 million euros, later frozen
- ChannelPhone calls appearing to come from the ministry
Business leaders including Giorgio Armani, Patrizio Bertelli of Prada, Diego Della Valle of Tod's and Marco Tronchetti Provera of Pirelli received calls that appeared to come from the defence ministry in Rome. A staff member put them through to a man speaking in the cloned voice of minister Guido Crosetto, who asked for around one million euros to free kidnapped Italian journalists, to be reimbursed by the Bank of Italy. Massimo Moratti, former owner of Inter Milan, paid. The money was traced to a Dutch bank account and frozen. Crosetto learned of the scam when a confused businessman called him.
Susie Wiles and Marco Rubio
- TargetForeign ministers, US governors, senators and executives
- OutcomeOngoing campaign, per the FBI
- ChannelSignal, text and voicemail
In May, senators, governors and executives received texts and calls from someone using the contact list of White House chief of staff Susie Wiles. Some who answered heard a voice that sounded like her. In July, a State Department cable disclosed that someone had used a cloned voice of Secretary of State Marco Rubio to leave Signal voicemails for at least three foreign ministers, a US governor and a US senator, alongside messages from a fake state.gov address. The FBI issued a public warning in May and updated it in December 2025, saying the campaign was ongoing and that AI had removed the usual tells.
Source: PBS NewsHour, July 2025 and FBI IC3 alert
Fideuram, Italy, 95 million euros
- TargetChairman of Fideuram, Intesa Sanpaolo's private bank
- Loss95 million euros out, about 36 million still missing
- ChannelWhatsApp, then phone call, then email
Paolo Molesini, chairman of Fideuram, the private banking arm of Intesa Sanpaolo, received a WhatsApp message that appeared to come from group chief executive Carlo Messina. A call followed in the cloned voice of a senior partner at a major law firm, who had no involvement. An email from a spoofed law firm domain supplied account details. Between 23 and 25 February, around 95 million euros left the bank in a series of transfers to accounts in China and Hong Kong, some routed through Portugal. Italian, Chinese and Portuguese authorities recovered more than half. About 36 million euros remained missing when Reuters reported the case in September 2026. Intesa has said it has blocked thousands of AI impersonation attempts aimed at its CEO since the start of the year. This is the largest confirmed loss from a voice clone to date.
Source: Cybernews, September 2026 and Gulf News, both citing Reuters and Corriere della Sera
Michigan, homebuyers, 66,000 dollars
- TargetA couple days before closing on a house
- Loss66,000 dollars
- ChannelEmail plus phone calls
A couple in Michigan lost their down payment in the days before closing. The attackers combined emails that mimicked their title company with phone calls that investigators believe used a cloned voice. Closings suit this fraud: the sum is large, the date is fixed, and the buyer and the title agent have often never met.
Source: CNN, September 2026
Netherlands, 170 victims in three weeks
- TargetBank customers across the Netherlands
- LossAbout 380,000 euros
- OutcomeTwo arrested
- ChannelAutomated phone calls
Rotterdam police arrested a 23 year old from Vaassen and a 25 year old from Ede. Their operation called people across the Netherlands with AI-generated recordings in the voices of bank employees, in dozens of variants covering different banks and scenarios, warning of suspicious activity. Victims were told to install an app that read their bank card over NFC, after which accomplices withdrew cash at ATMs. 170 people lost around 380,000 euros in three weeks. Most were in their twenties and thirties. Dutch banks spotted the method and raised it with the Electronic Crimes Task Force, which led to the arrests.
Source: Security Affairs, October 2026
How much money is lost to voice deepfakes?
No agency publishes a complete figure, and most numbers circulating online have no source. The best audited data comes from the FBI's Internet Crime Complaint Center. Its report for 2025, the first with a dedicated AI section, records 22,364 complaints that involved AI and 893 million dollars in losses. People over 60 accounted for 39 percent of that. The subset the FBI ties to cloned voices, the family distress scams, came to just over five million dollars. Business email compromise, where attackers now add a cloned voice to confirm a fraudulent wire, reached 3.05 billion dollars in reported losses the same year. Congressional researchers estimate fewer than 5 percent of voice clone victims report at all.
Source: FBI IC3 2025 Internet Crime Report