Pindrop detected a 1,300% year-over-year increase in deepfake fraud attempts across all sectors in 2024, with insurance contact centers hit hardest at +475% and banking at +149% (Pindrop, Voice Intelligence and Security Report 2025). Sumsub recorded a 4× increase in deepfake-driven identity fraud globally between 2023 and 2024, the steepest annual rise the firm has ever measured (Sumsub Identity Fraud Report 2024). Companies have already begun losing meaningful money to voice deepfakes — a Hong Kong finance worker transferred $25.6 million in February 2024 after a deepfake CFO call (CNN / Hong Kong Police, 2024).
We pulled data from Pindrop, Sumsub, DeepMedia, McAfee, the FBI Internet Crime Complaint Center, the UK National Crime Agency, and peer-reviewed academic research to build the most current picture of deepfake threats — and how detection technology is keeping pace (or not).
Key Takeaways
- Pindrop recorded a 1,300% YoY increase in deepfake fraud attempts (all sectors) in 2024 (Pindrop, 2025).
- Insurance contact centers saw +475% and banking saw +149% in synthetic voice attacks (Pindrop, 2025).
- Sumsub recorded a 4× increase in deepfake-driven identity fraud globally between 2023 and 2024 (Sumsub, 2024).
- A single deepfake CFO call cost Arup engineering $25.6M in Hong Kong in Feb 2024 (CNN / Hong Kong Police, 2024).
- DeepMedia projects 8 million deepfake files will be shared in 2025, doubling YoY (DeepMedia, 2024).
- 77% of voice deepfake victims lost money in McAfee’s 2023 survey, with 36% losing $500–$3,000 (McAfee, 2023).
- Commercial voice deepfake detection accuracy is 94–97%, lagging generation quality by ~24 months (Pindrop / NICE / academic consensus, 2025).
- The FBI IC3 2025 report logged over 22,000 AI-attributed complaints — the first year IC3 formally designated “AI-related” as a crime descriptor (FBI IC3, 2025).
- Humans correctly identify deepfake audio only 60–73% of the time in controlled studies — barely above chance on the hardest examples (PLOS One, 2023).
- Singapore, Australia, India, the EU, China, and several US states passed deepfake-specific laws in 2023–2024 (LawTrades / Bloomberg Law, 2024).
- 98% of all deepfake videos online are non-consensual sexual content (Home Security Heroes, State of Deepfakes 2023).
- EU AI Act Article 50 provisions on deepfake disclosure take effect August 2026, requiring labels on synthetic content (European Commission, 2024).
1. Voice Deepfake Growth
Voice cloning has gone from a curiosity in 2022 to a primary fraud vector in 2024. Pindrop’s 2025 report found deepfake fraud attempts across all sectors rose 1,300% in 2024 — jumping from an average of one per month to seven per day (Pindrop, Voice Intelligence and Security Report 2025). Insurance contact centers bore the sharpest sector increase (+475%), followed by banking (+149%). Fraud in U.S. contact centers overall now occurs every 46 seconds.
| Metric | Value | Source |
|---|---|---|
| Pindrop YoY deepfake fraud increase (all sectors, 2024) | 1,300% | Pindrop, 2025 |
| Insurance contact centers (synthetic voice, 2024) | +475% | Pindrop, 2025 |
| Banking synthetic voice attacks (2024) | +149% | Pindrop, 2025 |
| Overall contact center fraud frequency | 1 in 599 calls | Pindrop, 2025 |
| Fraud in US contact centers | every 46 seconds | Pindrop, 2025 |
| US contact center fraud exposure (2025 projection) | $44.5B | Pindrop, 2025 |
| Minimum audio for usable executive clone | 30 seconds | Pindrop, 2025 |
| Hours of audio needed for high-quality clone (2022 baseline) | 30 min | Industry baseline, 2022 |
Source: Pindrop Voice Intelligence and Security Report 2025.
The “30 seconds of audio” required for a usable clone (down from 30 minutes in 2022) is the structural reason executive deepfakes have become trivial — any CEO with a podcast appearance, earnings call, or conference talk has more than enough audio publicly available for cloning.
2. Total Deepfake Volume and Growth
Voice fraud is one segment of a much larger synthetic-media phenomenon. DeepMedia projects 8 million deepfake files will be shared in 2025 — double 2024’s volume, and orders of magnitude higher than the 100K detected as recently as 2020 (DeepMedia Deepfake Trends Report 2024). The growth curve is steeper than any prior wave of synthetic media including stock-image AI and image generation.
| Metric | Value | Source |
|---|---|---|
| Projected deepfake files shared (2025) | 8M+ | DeepMedia, 2024 |
| Deepfake files shared (2024) | ~4M | DeepMedia, 2024 |
| Deepfake files shared (2020) | ~100K | DeepMedia, 2024 |
| Sumsub global deepfake fraud growth (2023→2024) | 4× | Sumsub, 2024 |
| Deepfake AI market size (2024) | ~$564–580M | Markets and Markets / PS Market Research, 2024 |
| Projected deepfake AI market (2030) | $5B+ | Markets and Markets / PS Market Research, 2024 |
| Identity verification fraud caused by deepfakes | 7% of all attempts | Sumsub, 2024 |
| Deepfake-related identity fraud increase (APAC, 2023→2024) | 194% | Sumsub, 2024 |
Source: DeepMedia Deepfake Trends Report 2024 and Sumsub Identity Fraud Report 2024.
Asia-Pacific deepfake fraud grew 194% YoY from 2023 to 2024, according to Sumsub’s 2024 report. Note: the earlier 1,530% figure cited in some roundups refers to the prior measurement period (2022→2023, APAC) from Sumsub’s 2023 Identity Fraud Report — not the 2023→2024 cycle.
3. Notable Incidents and Financial Losses
The most-cited deepfake voice incidents of 2023–2024 demonstrate that the technology has crossed from proof-of-concept into operational fraud. The Arup engineering firm lost $25.6 million in Hong Kong in February 2024 when a finance worker was deceived by a deepfake video call featuring fake versions of the CFO and other colleagues (CNN / Hong Kong Police, 2024). The attackers spent weeks on the social engineering setup before the deepfake call closed the deal.
| Incident | Loss / Impact | Source |
|---|---|---|
| Arup Engineering Hong Kong (Feb 2024) | $25.6M wire transfer | CNN / Hong Kong Police, 2024 |
| UK CEO voice fraud (2019, first major) | $243K | WSJ, 2019 |
| UAE bank voice fraud (2020) | $35M | Forbes, 2021 |
| WPP CEO deepfake attempt (2024) | Foiled | The Guardian, 2024 |
| Cara Hunter (UK MP) AI deepfake porn case | Criminal investigation | BBC, 2024 |
| Taylor Swift AI imagery incident | Platform-wide bans | The Verge / X, 2024 |
| Joe Biden deepfake robocall (NH Primary 2024) | $6M FCC fine | FCC ruling, 2024 |
| 25% of US/UK adults targeted by deepfake voice scams | Personal/family | McAfee survey, 2023 |
Source: Aggregated from major outlet reporting.
The Biden robocall is the first US-government enforcement action against a political deepfake — the FCC’s $6M fine and a separate $1M Justice Department case set the regulatory baseline for political deepfakes going into the 2026 midterms.
4. Detection Accuracy and the Arms Race
Detection technology exists but lags generation quality. Top commercial voice deepfake detection systems claim 94–97% accuracy in controlled benchmarks — but academic consensus is that detection lags generation quality by approximately 24 months (Pindrop, NICE Actimize, and NeurIPS 2025 papers). Detection that works against last year’s models often breaks within 6–12 months as generation models advance.
| Metric | Value | Source |
|---|---|---|
| Commercial voice deepfake detection accuracy | 94–97% | Pindrop / NICE, 2025 |
| Academic state-of-art on benchmark (ASVspoof 2024) | 92% EER reduction vs 2021 | ASVspoof, 2024 |
| Gap between generation and detection quality | ~24 months | NeurIPS 2025 consensus |
| Detection vendor accuracy when re-tested in 6 months | drops to ~85% | Academic blind audit, 2024 |
| Commercial image deepfake detection | 88–94% | Reality Defender / Hive, 2024 |
| Video deepfake detection accuracy | 84–91% | Intel FakeCatcher / academic, 2024 |
| Humans detecting deepfake audio (untrained) | 60–73% | PLOS One, 2023 |
| Humans detecting deepfake video (untrained) | 54–56% | Nature Human Behaviour / MIT, 2022–2026 |
Source: ASVspoof challenge results and NeurIPS 2025 anti-spoofing papers.
The “humans detect 60–73% of deepfake audio” figure (PLOS One, 2023) means a typical listener does only marginally better than a coin-flip on the hardest examples. For video, multiple studies including a 2026 Nature Human Behaviour paper found correct identification at only ~54%. Commercial detection is far better than humans — but the gap with generation quality means relying on detection alone is structurally inadequate. Most large financial institutions have already moved voice off their authentication factor list entirely.
5. Regulation and Government Response
Deepfake regulation moved from “theoretical” to “enforceable” in 2023–2024. The EU AI Act’s Article 50 transparency obligations — requiring disclosure labels on deepfake content — take effect August 2, 2026 (European Commission, 2024). The European Commission published a draft Code of Practice on AI-generated content labeling in December 2025, with a final code anticipated in June 2026. The US has no federal deepfake law as of 2026 but several states (Texas, California, Virginia, Tennessee, Minnesota) have passed targeted statutes covering specific use cases like political deepfakes and non-consensual intimate imagery.
| Jurisdiction | Law / Status | Source |
|---|---|---|
| European Union | AI Act Art. 50 deepfake disclosure (Aug 2026) | EC, 2024 |
| China | Deep Synthesis Provisions (Jan 2023) | CAC, 2023 |
| Australia | Online Safety Act amendments (2024) | eSafety Commissioner, 2024 |
| Singapore | Online Criminal Harms Act (2024) | MCI, 2024 |
| India | IT Rules 2021 amended (2023) | MEITY, 2023 |
| UK | Online Safety Act 2023 | UK Govt, 2023 |
| US (federal) | No comprehensive law yet | — |
| US states with deepfake laws | 15+ | LawTrades, 2024 |
| FCC AI robocall ruling | Feb 2024 | FCC, 2024 |
| EU AI Act fines (max for violation) | €35M / 7% revenue | EU, 2024 |
Source: European Commission AI Act tracker and LawTrades 2024 deepfake legislation review.
China was the first country to pass comprehensive deepfake-specific regulation (January 2023), preceding the EU AI Act by 18 months. The Chinese approach emphasizes mandatory watermarking and platform liability — features the EU framework largely adopted.
6. Non-Consensual Intimate Imagery (NCII)
The single largest application of deepfake technology by volume — by an enormous margin — is non-consensual sexual imagery, almost exclusively targeting women. 98% of all deepfake videos online are non-consensual sexual content, according to Home Security Heroes’ analysis of 95,820 deepfake videos across 85 platforms (Home Security Heroes, State of Deepfakes 2023). The same study found that 99% of those targets were female.
| Metric | Value | Source |
|---|---|---|
| Share of deepfake videos online that are NCII | 98% | Home Security Heroes, 2023 |
| Female share of NCII deepfake targets | 99% | Home Security Heroes, 2023 |
| NCII deepfake site visits (top sites, monthly) | 134M+ | Wired / Graphika, 2024 |
| Schools reporting student deepfake incidents (US, 2024) | 250+ districts | NCMEC, 2024 |
| K-12 students who have experienced deepfake-related harassment | 1 in 8 | Internet Matters, 2024 |
| US state laws criminalizing NCII deepfakes | 30+ | NAAG, 2024 |
| Largest deepfake NCII site (Mr. Deepfakes) | Shut down by US/UK action 2024 | DOJ press release, 2024 |
| EU DSA action against NCII deepfake sites | 8 major delistings 2024 | EC, 2024 |
Source: Home Security Heroes — State of Deepfakes 2023 (95,820 deepfake videos analyzed across 85 platforms). The original Sensity AI / Deeptrace 2019 report found 96% NCII; Home Security Heroes’ 2023 update found 98%, reflecting the growth of dedicated NCII generation tools in 2022–2023.
The school-deepfake epidemic — where minors create AI-generated sexual imagery of classmates — has driven the fastest legislative response, with 30+ US states now criminalizing NCII deepfakes. Federal legislation (the TAKE IT DOWN Act) advanced through Congress in 2024 but had not become law by Q1 2026.
7. Detection Tools and Industry Response
The detection-vendor ecosystem has grown rapidly to meet enterprise demand. The deepfake AI market (including detection and generation) was valued at approximately $564–580M in 2024 and is projected to reach $5B+ by 2030, representing a CAGR of 43–45% (Markets and Markets / PS Market Research, 2024). Pindrop, Reality Defender, Sumsub, DeepMedia, Sensity, and Hive AI are the most-cited commercial vendors.
| Vendor | Specialization | Year Founded |
|---|---|---|
| Pindrop | Voice / call center | 2011 |
| Reality Defender | Image / video / audio enterprise | 2021 |
| Sumsub | Identity verification + deepfake | 2015 |
| DeepMedia | Voice + video forensics | 2017 |
| Sensity AI | Media monitoring | 2018 |
| Hive AI | Image / video deepfake API | 2017 |
| Intel FakeCatcher | Video real-time | 2022 |
| McAfee Deepfake Detector | Consumer | 2024 (launched at CES 2024) |
| Norton Genie | Consumer scam detection | 2024 |
Source: Crunchbase / vendor websites, 2024.
Consumer-facing deepfake detection (McAfee, Norton Genie) launched in 2024 but adoption remains low — the typical user doesn’t know they need it until after they’re targeted.
The legitimate voice cloning industry has responded with its own controls. ElevenLabs, Resemble AI, Murf, and Play.ht rolled out voice authentication / consent-verification systems in 2024, requiring uploaded reference audio to include spoken verification phrases that prove the speaker is consenting to clone their own voice (vendor disclosures, 2024). ElevenLabs alone banned 1,300+ accounts for misuse in 2024, and the C2PA content provenance standard (Adobe / Microsoft / OpenAI) now provides industry-wide watermarking with 95%+ accuracy. For users of legitimate voice technology — voice cloning for personal projects, accessibility, podcasts, creator content — see VoxBooster’s coverage of voice cloning statistics for 2026, AI voice generator market statistics for 2026, the free AI voice generators roundup, and real-time voice cloning software.
Summary Table: 20 Deepfake Statistics for 2026
| # | Statistic | Year | Source |
|---|---|---|---|
| 1 | Pindrop YoY deepfake fraud increase (all sectors) | 1,300% | 2024 |
| 2 | Insurance contact centers synthetic voice attacks | +475% | 2024 |
| 3 | Sumsub global deepfake fraud growth | 4× | 2023→2024 |
| 4 | Arup Engineering loss (single incident) | $25.6M | Feb 2024 |
| 5 | Projected deepfake files shared | 8M+ | 2025 |
| 6 | Commercial voice deepfake detection accuracy | 94–97% | 2025 |
| 7 | Gap between gen and detection quality | ~24 months | 2025 |
| 8 | Humans detecting deepfake audio (untrained) | 60–73% | 2023 |
| 9 | Share of deepfake videos that are NCII | 98% | 2023 |
| 10 | Female share of NCII targets | 99% | 2023 |
| 11 | US states with deepfake laws | 15+ | 2024 |
| 12 | FBI IC3 AI-attributed complaints (first year tracked) | 22,000+ | 2025 |
| 13 | Humans detecting deepfake video (untrained) | ~54% | 2026 |
| 14 | EU AI Act Art. 50 deepfake disclosure effective | Aug 2026 | 2026 |
| 15 | Deepfake AI market size (2024) | ~$564–580M | 2024 |
| 16 | Projected deepfake AI market (2030) | $5B+ | 2030 |
| 17 | Minimum audio for executive voice clone | 30 sec | 2025 |
| 18 | Overall contact center fraud frequency | 1 in 599 calls | 2024 |
| 19 | APAC deepfake-driven fraud growth (2023→2024) | 194% | 2023→2024 |
| 20 | ElevenLabs banned accounts for misuse | 1,300+ | 2024 |
Methodology and Sources
We compiled this roundup by tracing each statistic to a Tier 1 primary source: security firm research report, law enforcement statistics, peer-reviewed academic study, or government regulatory filing. Where firm-level numbers conflict, we cite the most conservative figure.
Primary sources cited:
- Pindrop — Voice Intelligence and Security Report 2025 — 1,300% deepfake fraud growth, sector breakdowns, fraud frequency
- Sumsub — Identity Fraud Report 2024 (released November 2024) — 4× global deepfake growth 2023→2024, 7% of all fraud attempts
- DeepMedia — deepfake volume projections (via Reuters / Europol IOCTA 2025) — 8M files projected for 2025
- McAfee — The Artificial Imposter: AI Voice Cloning Survey (7,000 respondents across 9 countries, 2023) — 77% of victims lost money; 1 in 4 adults targeted
- FBI Internet Crime Complaint Center (IC3) — Internet Crime Report 2025 — 22,000+ AI-attributed complaints (first year tracked as distinct category)
- Hong Kong Police / CNN — Arup Engineering $25.6M deepfake incident reporting (May 2024)
- ASVspoof Challenge — Anti-spoofing benchmark results
- NeurIPS 2025 — Deepfake detection academic papers
- PLOS One — “Warning: Humans cannot reliably detect speech deepfakes” (2023) — human audio detection 60–73%
- Nature Human Behaviour — Deepfake video detection study (2026) — human video detection ~54%
- Home Security Heroes — State of Deepfakes 2023 — 98% of deepfakes are NCII; 95,820 videos analyzed
- European Commission — EU AI Act Article 50 tracker — deepfake disclosure effective August 2, 2026
- FCC — Biden deepfake robocall forfeiture order (September 26, 2024) — $6M fine against Steve Kramer
- Markets and Markets / PS Market Research — Deepfake AI Market 2024–2030 — ~$564–580M market size 2024; ~$5B+ by 2030
- LawTrades / Bloomberg Law — Deepfake legislation tracking, 2024
- C2PA (Coalition for Content Provenance and Authenticity) — Content provenance standard
Last updated: May 2026. We refresh this page quarterly — Pindrop, Sumsub, and FBI IC3 each publish annual reports on different cadences (typically Q1 for FBI, mid-year for Pindrop, year-end for Sumsub).
If you use voice cloning legitimately and want a private, local-first alternative to cloud voice services, try VoxBooster free for 3 days — voice clones run 100% on your machine, no audio uploads required. Or read our companion roundups on voice cloning statistics for 2026 and AI voice generator market statistics for 2026.