Skip to content

Deepfake Statistics 2026: AI Voice Cloning, Deepfake Fraud and Detection

Deepfake statistics for 2026 from the FBI, FinCEN, Deloitte, Entrust, Sumsub, Regula, Gartner and peer-reviewed studies: fraud losses, voice cloning, KYC and detection.

Azad KhanBy , Senior GRC & Compliance18 min read

AI & LLM Penetration TestingFrom $4,500Fixed scope, free retest, no hourly billing.

Get my fixed price

This page collects the measured numbers on deepfakes: how many are reported, what they cost, how voice cloning feeds phone fraud, how often deepfakes reach identity verification, how well people and detectors spot them, and which rules now apply. The sources are the FBI’s Internet Crime Complaint Center (IC3), FinCEN, the FCC, the EU AI Act, Deloitte, Gartner, the identity verification firms Entrust, Sumsub, Signicat and Regula, and peer-reviewed studies. Each figure links to the organization that produced it and states the year it describes.

How to cite: link to this page or to the primary source beside each figure. Survey figures carry their sample size.

Key numbers at a glance

  • 22,364 complaints to the FBI in 2025 contained AI-related information, with adjusted losses of $893 million. (FBI IC3 2025 Internet Crime Report)
  • HK$200 million (about £20 million) was sent to criminals in 15 transfers after an Arup employee in Hong Kong joined a video call with deepfaked senior officers. (The Guardian, May 2024, reporting Arup’s and Hong Kong police statements)
  • 62% of organizations experienced a deepfake attack in the previous 12 months. (Gartner survey, September 2025, 302 cybersecurity leaders)
  • 1 in 5 biometric fraud attempts involved a deepfake, and injection attacks rose 40% year over year. (Entrust 2026 Identity Fraud Report, more than 1 billion identity verifications)
  • $40 billion: Deloitte’s forecast for US fraud losses enabled by generative AI by 2027, up from $12.3 billion in 2023. (Deloitte Center for Financial Services, 2024)
  • 73%: how often listeners correctly spotted speech deepfakes in a controlled study. (Mai et al., PLOS ONE, 2023, 529 participants)
  • 8 million deepfakes are estimated to have been shared in 2025, up from 500,000 in 2023. (UK Government, February 2026)

1. How many deepfakes there are

No agency counts every deepfake. These figures come from a government estimate, incident trackers built from news reports, and platform detections; each measures a different slice.

  • 8,000,000 deepfakes are estimated to have been shared in 2025, up from 500,000 in 2023, a sixteenfold rise by simple arithmetic. The announcement does not explain how the estimate was made. (UK Government, 2026)
  • 1,567 unique verified deepfake incidents were documented in 2025, deduplicated from 3,253 reported incidents. 20% involved child sexual abuse material or non-consensual intimate imagery. (Resemble AI 2025 Deepfake Threat Report)
  • 821 verified deepfake attacks were documented in the first half of 2026, drawn from 1,760 news reports, with at least 15,736 documented victims. About 1 in 6 (137 incidents) were non-consensual sexual imagery of adults or children. (Resemble AI H1 2026 Deepfake Threat Report)
  • Deepfake fraud attempts rose 94% in the UK in 2025, behind France (96%) and ahead of Spain (84%) and Germany (53%). (Sumsub Identity Fraud Report 2025-2026, 4,000,000+ fraud attempts analyzed)
  • Deepfake activity on phone calls rose 680% year over year in 2024. (Pindrop 2025 Voice Intelligence and Security Report, more than 1.2 billion calls)

A caution on one popular number: Europol updated its deepfake report in January 2024 and removed a statement on the expected future share of synthetically generated online content, saying it came from an inaccurate source. (Europol Innovation Lab, Facing Reality? 2022, updated 2024) Check any forecast of the “share of synthetic content online” against that correction before repeating it.

2. Deepfake fraud losses

  • 22,364 IC3 complaints in 2025 reported AI-related information, with adjusted losses of $893,346,472. IC3 notes that AI-enabled synthetic content “is becoming increasingly difficult to detect and easier to make.” (FBI IC3 2025 Internet Crime Report)
  • Investment complaints with a reported AI nexus lost $632,041,188 in 2025, while total investment scam losses exceeded $8 billion. IC3 reads the gap as a sign that many victims do not realize AI was involved. (FBI IC3 2025)
  • Businesses reported $30,256,592 in losses to business email compromise involving AI in 2025. (FBI IC3 2025)
  • Confidence and romance scams with a likely AI nexus cost victims $19,041,653 in 2025; “distress” scams that use voice cloning to imitate a relative cost more than $5 million. (FBI IC3 2025)
  • AI-involved employment scams, including voice deepfakes in online job interviews, cost $12,550,185 in 2025. (FBI IC3 2025)
  • $1.28 billion in deepfake-related losses was reported across 2025’s verified incidents, and more than 80% of incidents disclosed no financial figure at all. (Resemble AI 2025 Deepfake Threat Report)
  • Businesses lost an average of $450,000 to deepfakes; financial services firms lost more than $603,000, and 10% of organizations lost more than $1 million. (Regula Deepfake Trends 2024, 575 business decision-makers surveyed by Sapio Research, August 2024)
  • Within finance, fintech companies averaged $637,000 in deepfake losses and traditional banks $570,000; 23% of financial-sector respondents lost more than $1 million. (Regula 2024)
  • Generative AI could enable US fraud losses of $40 billion by 2027, up from $12.3 billion in 2023, a compound annual growth rate of 32%. Deloitte puts generative AI email fraud alone at about $11.5 billion by 2027 in its aggressive scenario. (Deloitte Center for Financial Services, May 2024)

The FBI figures are complaints in which AI was identified, so they are a floor. The incident trackers say the same thing from another angle: most victims never state a loss.

3. Executive impersonation and the Arup case

  • HK$200 million (£20 million) was transferred by an Arup employee in Hong Kong in January 2024 after video conference calls with people “posing as senior officers of the company.” Hong Kong police said the money went to five local bank accounts in 15 transactions. Arup confirmed that fake voices and images were used and that none of its internal systems were compromised. (The Guardian, May 17, 2024) Deloitte describes the same case as roughly US$25 million. (Deloitte, 2024)
  • The WPP chief executive was impersonated in a Microsoft Teams meeting set up from a fake WhatsApp account, using an AI voice clone and YouTube footage. The attempt targeted an agency leader and was unsuccessful. (The Guardian, May 10, 2024)
  • A LastPass employee received WhatsApp calls, texts and a voicemail carrying an audio deepfake of the company’s CEO in April 2024. The employee ignored the messages and reported them; the company said there was no impact. (LastPass, April 2024)
  • 25.9% of executives said their organization had experienced one or more deepfake incidents targeting financial and accounting data in the previous 12 months (15.1% one, 10.8% more than one). (Deloitte poll, May 2024, 2,190 executive votes)
  • 51.6% of executives expected the number and size of deepfake attacks on financial and accounting data to increase over the next 12 months. (Deloitte poll, 2024, 2,131 votes)
  • 33% of IT and security leaders named employee account takeover or impersonation as a workforce identity threat, and 46% do not measure AI-driven impersonation such as deepfakes at all. (Entrust KYE and Workforce Identity Crisis Report, October 2026, 1,002 US and UK IT and security leaders)
  • Since April 2025, actors have impersonated senior US officials with text messages and voice memos that may use AI-generated voices, to build rapport before trying to access personal accounts. (FBI PSA I-051525-PSA, May 2025)
  • Europol reports that deepfakes replicating voices, images and video “are now being used in CEO fraud.” (Europol EU-SOCTA 2025)

Each of these cases ended at a payment or access request. The control that stops them is a call-back or second approval on a channel the attacker does not control, and the only way to know staff use it is to test it. A phishing and social engineering test can include a voice pretext against your finance team; a red team engagement can chain it to a real payment workflow. Most of these attacks are spear phishing with a new delivery method.

See what a real pentest finds. Read our sample report, then get a fixed quote in writing.

4. Voice cloning and vishing

  • Voice phishing (vishing) rose 442% between the first and second half of 2024, driven by AI-assisted phishing and impersonation. (CrowdStrike 2025 Global Threat Report)
  • Synthetic voice fraud in insurance rose 475% in 2024, and overall fraud attempts on calls rose 26%. (Pindrop 2025 Voice Intelligence and Security Report, 1.2 billion calls)
  • 53% of fraudsters passed knowledge-based authentication checks and 1 in 4 passed one-time passcode challenges; caller ID spoofing appeared in over 16% of confirmed fraud cases. (Pindrop 2025)
  • 28% of UK adults say they were targeted by an AI voice cloning scam at least once in the past year, and 46% had never heard of such scams. 8% said they would send whatever was asked in a cloned-voice call from a family member, even if the call seemed strange. (Starling Bank, September 2024, 3,010 UK adults surveyed by Mortar Research)
  • 4 of the 6 popular voice cloning products tested failed to take basic steps to stop someone cloning a voice without consent. (Consumer Reports, March 2025)
  • Victims claimed more than $5 million in losses to distress scams in 2025, in which voice cloning imitates a loved one in trouble. (FBI IC3 2025)
  • The FBI advises families to agree on a secret word or phrase and to hang up and call back on a known number. (FBI PSA I-120324-PSA, December 2024)

The phone channel is where a cloned voice does its work, and the help desk and finance line are the usual targets. Our guide to vishing covers how these calls are run and the controls that hold up, and our phishing attack statistics put vishing beside email and SMS lures.

5. Deepfake identity and KYC fraud

  • 1 in 5 biometric fraud attempts involved a deepfake; deepfaked selfies rose 58% in 2025 and injection attacks rose 40% year over year. (Entrust 2026 Identity Fraud Report, over 1 billion verifications across 195 countries, data September 2024 to September 2025)
  • Digital forgeries made up 35% of document fraud in 2025, up from a 29% average in 2022 to 2024; national ID cards were 46% of fraudulent document submissions. (Entrust 2026)
  • A deepfake attempt occurred every five minutes in 2024, and digital document forgeries rose 244% year over year. (Entrust 2025 Identity Fraud Report, data September 2023 to August 2024)
  • “Sophisticated fraud,” meaning multi-step attacks that combine several techniques in one verification attempt, rose 180% worldwide in 2025. Mainstream AI tools such as ChatGPT and Gemini were behind 2% of all falsified documents Sumsub processed. (Sumsub Identity Fraud Report 2025-2026)
  • Deepfake attempts grew 2,137% over three years and now make up about 6.5% of fraud attempts Signicat detects, or 1 in 15; 42.5% of fraud attempts detected in the financial sector are attributed to AI. (Signicat, The Battle Against AI-Driven Identity Fraud, February 2025, 1,206 fraud decision-makers in seven European countries)
  • Injection attacks on identity verification increased 200% in 2023. (Gartner, February 2024)
  • FinCEN observed an increase, beginning in 2023 and continuing in 2024, in suspicious activity reports describing suspected deepfake media in fraud schemes, and asked banks to tag such reports “FIN-2024-DEEPFAKEFRAUD.” Its red flags include webcam plugins during live checks and photos that match galleries of AI-generated faces. (FinCEN Alert FIN-2024-Alert004, November 2024)

Hiring is now an identity check too:

  • 6% of job candidates admitted to interview fraud, either posing as someone else or having someone pose as them. Gartner predicts that by 2028, one in four candidate profiles worldwide will be fake. (Gartner, July 2025, 3,000 candidates surveyed in Q2 2025)
  • 28% of IT and security leaders have experienced or investigated candidate fraud in the past 12 to 24 months. (Entrust KYE Report, 2026, 1,002 respondents)
  • KnowBe4 hired a software engineer whose application photo began as stock photography and was “enhanced” with AI, after four video interviews. The workstation KnowBe4 sent began loading malware the moment it was received; the activity started at 9:55 p.m. and the device was contained around 10:20 p.m., with no data lost. (KnowBe4, July 2024)

6. Can people tell a deepfake from the real thing?

  • People shown AI-generated and real faces were 48.2% accurate, close to chance. With training and feedback, accuracy rose only to 59.0%. Synthetic faces were rated 7.7% more trustworthy than real ones. (Nightingale and Farid, PNAS, 2022, 315 and 219 participants)
  • Listeners detected speech deepfakes 73% of the time in English and Mandarin; showing them examples beforehand improved results only slightly. (Mai et al., PLOS ONE, 2023, 529 participants)
  • In a representative study of 3,002 participants in the US, Germany and China, most people were simply guessing when asked whether audio, images and text were human- or machine-made. (Frank et al., IEEE Symposium on Security and Privacy, 2024)
  • Ordinary people and the leading detection model were similarly accurate on deepfake videos but made different mistakes. People shown the model’s prediction did better than either alone, but wrong predictions often made them less accurate. (Groh et al., PNAS, 2022, 15,016 participants)
  • Just 0.1% of consumers correctly identified every real and deepfake image and video, even when told to look for deepfakes. Participants were 36% less likely to spot a synthetic video than a synthetic image, yet more than 60% were confident in their answers. (iProov, February 2025, 2,000 UK and US consumers)
  • 56% of businesses said they were very confident in their ability to detect deepfakes, but only 6% reported avoiding financial losses from deepfake attacks. (Regula 2024, 575 decision-makers)

Across faces, voices and video, training helps a little and confidence runs ahead of accuracy. “Spot the fake” is not a control; procedures that never rely on recognizing a face or voice are.

7. Detection tools and what organizations are doing

  • On Deepfake-Eval-2024, a benchmark of deepfakes that circulated in 2024 (45 hours of video, 56.5 hours of audio and 1,975 images from 88 websites in 52 languages), open-source detectors lost 50% of their AUC on video, 48% on audio and 45% on images compared with previous benchmarks. Commercial and fine-tuned models did better but did not yet match deepfake forensic analysts. (Chandra et al., Deepfake-Eval-2024, arXiv 2025)
  • By 2026, Gartner predicted, attacks using deepfakes on face biometrics would lead 30% of enterprises to stop treating identity verification and authentication solutions as reliable in isolation. (Gartner, February 2024)
  • Asked how they prepare staff for deepfake fraud, 22.1% of executives named communications about new threats, 19.0% training including role play, 10.8% new policies, 7.4% new detection technology, and 9.9% said they had taken no action. (Deloitte poll, 2024, 1,527 votes)
  • 62% of organizations experienced a deepfake attack involving social engineering or exploiting automated processes, and 32% faced an attack on AI applications through the application prompt. 67% of cybersecurity leaders said emerging generative AI risks demand significant changes to existing approaches. (Gartner, September 2025, 302 cybersecurity leaders surveyed March to May 2025)
  • Only 28% of fully remote organizations measure and regularly report AI-driven impersonation attacks, against 40% of hybrid organizations. (Entrust KYE Report, 2026, 1,002 respondents)

Detection tools are one layer. The controls that worked in the cases above were procedural: a second channel, refusing requests outside normal tools, and limits on what one person can approve. The phishing testing pricing page shows what a voice and email campaign covers, and our cybersecurity statistics page puts these numbers beside other attack types.

8. Deepfake laws and regulation

  • February 8, 2024: the FCC unanimously ruled that AI-generated voices are “artificial” under the Telephone Consumer Protection Act, making voice-cloned robocalls illegal without prior consent. A coalition of 26 state attorneys general had supported the approach, and the FCC has robocall agreements with 48 state attorneys general. (FCC news release, 2024; Declaratory Ruling FCC 24-17)
  • $6 million: the FCC fine adopted on September 26, 2024 against political consultant Steve Kramer for robocalls using a deepfake of President Biden’s voice before the New Hampshire primary. (FCC, 2024)
  • $1 million: the civil penalty Lingo Telecom agreed to pay for transmitting those calls, with new know-your-customer terms. (FCC, August 2024)
  • August 2, 2026: the date the EU AI Act’s Article 50 transparency obligations apply. Providers must mark synthetic audio, image, video and text output in a machine-readable format, and deployers must disclose that deepfake content has been artificially generated or manipulated. (European Commission, 2026; Regulation (EU) 2024/1689, Article 50)
  • €15 million or 3% of worldwide annual turnover, whichever is higher: the maximum fine for breaching Article 50 transparency obligations. (Regulation (EU) 2024/1689, Article 99(4))
  • 48 hours: the time covered platforms have to remove non-consensual intimate photos or videos, and known identical copies, after a valid request under the TAKE IT DOWN Act. The FTC began enforcing the platform duties on May 19, 2026, after its chairman wrote to 15 major platforms about compliance. (FTC, May 2026)
  • Two-thirds of US states had laws protecting the general public from AI-generated intimate deepfakes as of May 2025, and four more protected only minors. (Public Citizen, May 2025)
  • State laws on election deepfakes date back to Texas in 2019; Public Citizen tracks enacted and pending bills in every state, last updated November 9, 2026. (Public Citizen election deepfake tracker)

For financial institutions the regulatory expectation is already concrete: FinCEN’s alert lists the red flags examiners will look for, and its SAR key term means deepfake reporting can be counted.

What to expect in 2027

These are sourced forecasts and trends drawn from the data above, not our own predictions.

  • $40 billion in US fraud losses enabled by generative AI by 2027 is Deloitte’s base forecast, growing 32% a year. (Deloitte, 2024)
  • One in four candidate profiles worldwide will be fake by 2028, Gartner predicts, which makes hiring and onboarding an identity verification problem. (Gartner, 2025)
  • 72% of EU companies expect more sophisticated attacks using AI, particularly deepfakes. (Sumsub 2025-2026)
  • Trend: Europol expects AI tools, including voice cloning and deepfakes, to increase business email compromise and romance fraud. (Europol EU-SOCTA 2025)
  • Trend: the platform counts cited on this page all rose in their latest year. Deepfaked selfies rose 58% and injection attacks 40% in Entrust’s data (Entrust 2026), and deepfake attempts in the UK, France and Spain rose 84% to 96% in 2025 (Sumsub 2025-2026).

Part of our cybersecurity statistics hub. See also phishing attack statistics and cyber insurance claims statistics.

Sources

Frequently asked questions

How much money is lost to deepfake fraud? The FBI counted 22,364 complaints with AI-related information in 2025 and $893 million in adjusted losses, $632 million of it in investment scams. Incident trackers reported $1.28 billion in deepfake losses in 2025 while noting that more than 80% of incidents disclosed no figure. Deloitte forecasts generative AI could enable $40 billion in US fraud losses by 2027.

What happened in the Arup deepfake case? In January 2024 an Arup employee in Hong Kong joined video calls with people posing as senior officers and transferred HK$200 million (about £20 million, or roughly US$25 million) to five bank accounts in 15 transactions. Arup confirmed fake voices and images were used and that its internal systems were not compromised.

Can people reliably spot a deepfake? No. People judged AI-generated faces with 48.2% accuracy, and only 59.0% after training. Listeners caught speech deepfakes 73% of the time. In a 2,000-person study, 0.1% identified every real and fake item correctly, while more than 60% felt confident.

How common are deepfakes in identity verification fraud? Entrust found deepfakes in one in five biometric fraud attempts in its 2026 report, with deepfaked selfies up 58% and injection attacks up 40%. Signicat puts deepfakes at about 6.5% of the fraud attempts it detects, up 2,137% in three years.

Are AI voice robocalls illegal in the US? Yes, without the called party’s prior consent. The FCC ruled on February 8, 2024 that AI-generated voices are “artificial” under the Telephone Consumer Protection Act, and later fined the consultant behind the deepfaked Biden robocalls $6 million.

What does the EU AI Act require for deepfakes? From August 2, 2026, Article 50 requires deployers to disclose that deepfake content was artificially generated or manipulated, and requires providers to mark synthetic output in a machine-readable format. Breaches can be fined up to €15 million or 3% of worldwide annual turnover.

Azad Khan

Written by

Azad Khan, Senior GRC & Compliance

26+ years in the field. Focus: GRC, ISO 27001 & PCI DSS consulting. Part of the team that scopes, tests, and reports for our clients. Full profile →

Find out what an attacker sees.

Tell us what to test and see your fixed price.

Prefer the full scoping questionnaire? 
Start the conversation

Get a quick proposal

We are online