Deepfake Statistics 2026: Verified Facts on Fraud, Detection, and Growth

Deepfake statistics you can actually cite: fraud losses, detection accuracy, market size, and regulation, with every number traced to its primary source.

Every deepfake statistic on this page was checked against its primary source before publication, and every number links to that source with its publication year and its scope. If a figure could not be traced to the organization that originally published it, we cut it. Several numbers that appear in almost every other roundup failed that test, and we explain why below.

Key deepfake statistics for 2026: the FBI logged 22,364 complaints referencing AI with $893 million in losses in 2025, Entrust recorded a deepfake attempt every five minutes on its identity platform in 2024, and people correctly identify deepfakes only 55.5 percent of the time. Every figure below links to its primary source.

Editor's Picks: The 10 Deepfake Statistics That Hold Up

1. $25.6 million was stolen from engineering firm Arup after an employee in Hong Kong joined a video call where every other participant was a deepfake, per Fortune reporting (January 2024). It remains the canonical corporate deepfake case. Use $25.6 million, the USD figure, when citing it.
2. A deepfake attack attempt occurred every 5 minutes on Entrust's identity verification platform in 2024, per the Entrust 2025 Identity Fraud Report (2024 data). Scope caveat: this is one vendor's own-platform telemetry, not an internet-wide rate.
3. 22,364 complaints referencing AI, with $893 million in reported losses, were logged by the FBI's Internet Crime Complaint Center in 2025, the first year IC3 tracked AI as a crime descriptor, per the FBI IC3 2025 Annual Report (2025 data, published April 2026). Treat it as a floor: it only counts complaints that mentioned AI.
4. $40 billion is Deloitte's projection for US generative-AI-enabled fraud losses by 2027, up from $12.3 billion in 2023, with a conservative scenario of $22 billion, per the Deloitte Center for Financial Services (2024). This is a scenario model, not a measurement.
5. Contact-center deepfake attempts rose more than 1,300% in 2024, from roughly one per month to seven per day across 1.2 billion analyzed calls, per Pindrop's 2025 Voice Intelligence and Security Report (2024 data, vendor telemetry).
6. Vishing (voice phishing) attacks grew 442% between the first and second half of 2024, and H1 2025 volume already exceeded all of 2024, per the CrowdStrike Global Threat Report (2025).
7. 487 documented deepfake incidents caused $347 million in losses in Q2 2025 alone, per the Resemble AI Q2 2025 Deepfake Security Report (2025). Scope caveat: a vendor tracker of publicly reported cases, so it undercounts the real total.
8. Humans detect deepfakes only 55.5% of the time overall, barely better than a coin flip, in a peer-reviewed meta-analysis of 86,155 participants published in Human Behavior and Emerging Technologies (2024).
9. 1 in 4 candidate profiles will be fake by 2028, per a Gartner prediction (2025). A prediction, not a current measurement.
10. NCMEC logged more than 50,000 reports of financially motivated sextortion in 2025, roughly 137 per day, up from about 36,000 in 2024, per the National Center for Missing and Exploited Children (2026 publication).

Deepfake Fraud and Scam Statistics

1. $632 million of the FBI's $893 million in AI-referencing losses came from investment fraud, the category where deepfake celebrity endorsements concentrate, per the FBI IC3 2025 Annual Report (2025 data, published 2026).
2. About $13 million in reported losses came from voice and video deepfakes used in fake job interviews, per the same IC3 report (2025 data). Again a floor: only complaints that named AI are counted.
3. Documented deepfake incidents nearly tripled within 2025: 163 incidents with more than $200 million in losses in Q1, then 487 incidents with $347 million in Q2, per Resemble AI's Q1 and Q2 2025 reports. Both figures count only publicly reported cases.
4. Video was the leading deepfake modality at 46% of documented incidents in Q1 2025, and celebrities and politicians made up 41% of victims, per Resemble AI (2025).
5. The Arup theft moved $25.6 million through 15 transfers in one day after a single video call staffed entirely by deepfaked colleagues, per Fortune (2024).
6. Digital document forgeries rose 244% year over year in 2024, and deepfakes accounted for 24% of fraud attempts against motion-based biometric verification, per the Entrust 2025 Identity Fraud Report (2024 data, own-platform telemetry).
7. Deepfakes accounted for 11% of first-party fraud methods, and multi-step fraud attacks rose 180% year over year, per the Sumsub Identity Fraud Report 2025-2026 (2025). Caveat: earlier Sumsub editions used a different denominator for this share, so do not blend figures across report years.

For how these schemes actually play out, and what to do if one targets your company, see our guide to deepfake scams.

Voice Cloning and Audio Deepfake Statistics

1. Synthetic voice attacks rose 475% at insurance companies and 149% at banks in 2024, from Pindrop's analysis of 1.2 billion customer calls, per the Pindrop 2025 Voice Intelligence and Security Report (2025).
2. Pindrop projected deepfake-related contact-center fraud to grow another 162% in 2025, per the same report. Note this is a forward projection, not a measured result.
3. Vishing grew 442% from H1 to H2 2024, and CrowdStrike observed more vishing in the first half of 2025 than in all of 2024, per the CrowdStrike Global Threat Report (2025).
4. 4 of 6 leading voice cloning tools had no meaningful safeguard against cloning a voice without consent, per a Consumer Reports assessment (March 2025).

For documented case files behind these numbers, see our guide to voice cloning scams.

Hiring Fraud and Identity Verification Statistics

1. Gartner predicts 1 in 4 candidate profiles will be fake by 2028, and 6% of candidates in its 2Q25 survey (n=3,000) admitted to interview fraud, per Gartner (2025).
2. 16.8% of applicants to one security vendor's own job openings turned out to be fake, per Pindrop's CEO in Fortune (2025). One company's hiring pipeline, but a rare measured data point.
3. Native virtual-camera injection attacks rose 2,665% year over year, and face-swap attacks rose 300% versus 2023, per the iProov Threat Intelligence Report (2025). Scope caveat: vendor telemetry from iProov's own identity-verification traffic.

Deepfake Detection Statistics: Humans vs Machines

1. Humans detect deepfakes with 55.5% overall accuracy, split by media type as audio 62.1%, video 57.3%, and image 53.2%, in a peer-reviewed meta-analysis covering 86,155 participants, published in Human Behavior and Emerging Technologies (2024). Image deepfakes are essentially a coin flip.
2. State-of-the-art open-source detectors lose roughly 45 to 50% of their AUC when moving from academic benchmarks to real-world, in-the-wild deepfakes, per the Deepfake-Eval-2024 benchmark study (2025). This is the industry's most important honesty check: lab accuracy does not survive contact with the wild.

Read those two findings together and the picture is clear: people cannot reliably spot deepfakes unaided, and no detector, ours included, is infallible on in-the-wild content. That is why every result should be treated as a confidence score, not a certainty, and why vendor claims of near-perfect accuracy deserve skepticism. We will publish audited first-party platform data in our quarterly State of Deepfakes report starting late 2026. Until then, pair automated checks with the manual cues in our guide on how to spot a deepfake.

Elections, Sextortion, and Consumer Protection

1. Only 27 viral AI-enabled disinformation campaigns were identified around the 2024 UK, French, and EU elections, with no evidence any election outcome was changed, per the Alan Turing Institute's CETaS research report (2024). The honest framing: deepfakes are a real electoral risk, but claims that they have already swung elections are not supported by evidence.
2. NCMEC received more than 50,000 reports of financially motivated sextortion in 2025, roughly 137 per day and up from about 36,000 in 2024, per NCMEC (2026 publication).
3. Generative-AI-related child exploitation reports to NCMEC jumped from 6,835 in the first half of 2024 to 440,419 in the first half of 2025, per NCMEC (2026 publication). Scope caveat: part of that jump reflects new platforms beginning to report, not purely new abuse.

Deepfake Detection Market Size and Forecast

There is no single agreed deepfake detection market size. Analyst firms disagree by an order of magnitude because they draw the market boundary differently: some count all "deepfake AI" including generation tools, others count detection only. Cite the firm and its definition, never a bare number.

The consistent signal across all three: the detection-focused market is small today (Market.us puts pure detection near $114 million in 2024) but every firm models compound annual growth above 40%. The disagreement is about scope, not direction.

Deepfake Regulation: Key Dates and State Law Counts

1. EU AI Act Article 50 transparency obligations apply from August 2, 2026: deployers must disclose deepfakes, and providers must make synthetic content machine-detectable, with fines up to 15 million euros or 3% of worldwide turnover, per Regulation (EU) 2024/1689 (official EUR-Lex text).
2. The US TAKE IT DOWN Act's platform obligations have been enforced since May 19, 2026: covered platforms must remove non-consensual intimate imagery, including AI-generated imagery, within 48 hours of a valid request, per S.146, 119th Congress (signed May 2025).
3. Roughly 30 US states have election-deepfake laws and 46 states have laws covering non-consensual intimate deepfakes, per Public Citizen's election deepfake tracker and intimate deepfake tracker (2026). Caveat: counts vary by methodology and change frequently, so check the trackers before citing an exact number.

How We Verify These Deepfake Statistics (and How to Cite Us)

Methodology. Every statistic above traces to the organization that originally published it: government data (FBI IC3, NCMEC, EUR-Lex, Congress.gov), named vendor research with disclosed methodology (Entrust, Sumsub, Pindrop, CrowdStrike, Resemble AI, iProov), peer-reviewed and preprint academic work (the Wiley meta-analysis, Deepfake-Eval-2024, the Alan Turing Institute), and named analyst firms with their market definitions stated. Where a figure is a projection, a survey, or one vendor's own-platform telemetry, we say so next to the number. We exclude figures that circulate only in secondary roundups. Statistics are re-verified quarterly, and the updated date at the top of this page changes with every review.

What we cut. Several famous deepfake numbers do not appear on this page because they fail sourcing: viral file-count projections from 2023 that were never measured, decade-old percentages about deepfake content categories still cited as current, and untraceable dollar totals for annual deepfake fraud. If a number is not here, that is usually why.

How to cite us. You are welcome to cite any statistic or embed any chart on this page with attribution:

Suggested citation: DeepfakeDetector.ai, "Deepfake Statistics 2026: Verified Facts on Fraud, Detection, and Growth," updated July 14, 2026, https://deepfakedetector.ai/blog/deepfake-statistics

The charts are original and free to republish with a link back to this page as the source. Journalists on deadline: if you need a number checked, email us and we will verify it against the primary source for you.

FAQ

How many deepfakes are on the internet? No verified count exists. The totals that circulate widely trace back to a single 2023 vendor projection that was never measured against reality, so we do not republish them. Treat any precise global deepfake count as unsourced.

How much money has been lost to deepfake fraud? There is no single global total. The clearest sourced markers: $893 million in losses across 22,364 AI-referencing complaints in the FBI's 2025 IC3 report, $347 million in publicly documented deepfake incidents in Q2 2025 per Resemble AI, and $25.6 million in the single Arup case in 2024.

What percentage of people can spot a deepfake? A peer-reviewed meta-analysis of 86,155 participants found 55.5% overall human detection accuracy: audio 62.1%, video 57.3%, and image 53.2%. Since 50% is pure chance, unaided human judgment is close to a coin flip.

How fast is deepfake fraud growing? The best-sourced growth rates: contact-center deepfake attempts rose more than 1,300% in 2024 (Pindrop), vishing grew 442% between the halves of 2024 (CrowdStrike), and Entrust logged a deepfake attempt every five minutes on its platform. Each is one organization's telemetry, not a global rate.

How often is this page updated? Quarterly. The updated date at the top of the page changes with every review, and stats that fail re-verification are removed or corrected.

Conclusion: Deepfake Statistics Worth Citing

These deepfake statistics tell one story: documented deepfake fraud is growing on every measured channel, humans detect fakes at little better than chance, and even automated detectors degrade on in-the-wild content. The honest response is layered verification, not blind trust in any single number or tool. Use any figure here with its source, year, and scope caveat, and if you want the deeper mechanics, start with our guide to deepfake detection.

Related reading