Deepfake Fraud Statistics 2026: Verified AI Fraud Data for Insurers
Sourced deepfake and AI fraud statistics for 2026: growth, losses, voice cloning, human detection rates and regulation, each linked to its original source.
This page collects statistics on AI-enabled and deepfake fraud that matter to insurers, each traced to the organization that published it. It complements our broader insurance fraud statistics 2026 page, which covers total fraud costs by country and line.
Last updated: September 29, 2026
A caution before the numbers. Almost all deepfake fraud data comes from identity-verification, voice-security and consulting firms. It covers financial services broadly, uses each vendor’s own customer base or survey sample, and is not insurance-specific. No industry body yet publishes a measured figure for deepfake fraud in insurance claims. We include each figure with its scope so you can judge how far it travels.
Growth and Prevalence
| Finding | Period | Source |
|---|---|---|
| Deepfakes detected worldwide rose 4x | 2023 → 2024 | Sumsub, Nov 2024 |
| Deepfakes detected rose 10x | 2022 → 2023 | Sumsub Identity Fraud Report 2023 |
| Deepfake attempts rose 3,000% (31x); digitally forged identities rose 5x | 2022 → 2023 | Onfido, via Entrust |
| Deepfakes rose 2,137% and now make up 6.5% of fraud attempts (about 1 in 15) | Three years to 2024 | Signicat, 2025 |
| Deepfake fraud attempts in contact centers rose more than 1,300% | 2024 | Pindrop, June 2025 |
- Identity fraud more than doubled between 2021 and 2024, from 1.10% to 2.50% of verifications. Deepfakes were one of the main trends of 2024, and banking and insurance was among the five most affected sectors (Sumsub Identity Fraud Report 2024).
- About half of businesses surveyed had experienced audio deepfake fraud, and about half video deepfake fraud. On a like-for-like comparison, 49% had experienced both kinds, up from 37% (audio) and 29% (video) in 2022 (Regula, Deepfake Trends 2024).
- In the same survey, 58% of businesses had been deceived by fake or modified documents, up from 49%, and 92% had experienced some form of identity fraud in the previous 12 months.
- Sumsub’s 2025–2026 report puts the global fraud rate at 2.2% of verifications, lower than the 2024 peak, and reports a 180% rise in sophisticated, multi-step fraud (Sumsub, November 2025).
Insurance-Specific Signals
Insurance-specific deepfake data is thin. These are the clearest published signals:
- Synthetic-voice attacks on insurers rose 475% in 2024, faster than in banking (+149%) or retail (+107%) (Pindrop 2025 Voice Intelligence and Security Report).
- Banking and insurance had an identity-fraud rate of 2.7% in 2024, third highest of the sectors Sumsub tracks, with 162% growth over 2023 (Sumsub, November 2024).
- The UK’s Association of British Insurers and the Insurance Council of Australia publish detected-fraud totals, but neither yet breaks out AI-generated evidence as a category (see our insurance fraud statistics page).
If you see a precise figure for “deepfake insurance fraud losses”, ask where it comes from. We could not find one with a published method.
Financial Losses
- US$40 billion by 2027. Deloitte’s Center for Financial Services projects that generative AI could enable US fraud losses of US$40 billion by 2027, up from US$12.3 billion in 2023, a 32% compound annual growth rate. The projection covers generative-AI-enabled fraud in financial services, not deepfakes alone (Deloitte, May 2024).
- About US$450,000 per business. Businesses surveyed by Regula lost an average of nearly US$450,000 each to deepfakes. Financial-services firms lost a little over US$600,000 and fintechs more than US$630,000. These are per-business averages, not per-incident costs (Regula, October 2024; Sapio Research survey of 575 decision-makers in Germany, Mexico, Singapore, the UAE and the US).
- US$12.5 billion in reported fraud losses. US consumers reported losing more than US$12.5 billion to fraud of all kinds in 2024, from 2.6 million consumer fraud reports (FTC).
- US$25.6 million in one video call. In early 2024, Hong Kong police disclosed that a finance worker at a multinational had transferred HK$200 million (about US$25.6 million) in 15 transactions after a video conference in which the other participants, including the “CFO”, were deepfakes (CNN, February 2024). The company was later confirmed as the engineering firm Arup.
Voice Cloning
- Three seconds of audio. Microsoft’s VALL-E research showed speech synthesis in a new speaker’s voice from a three-second enrolled recording (Wang et al., 2023). Commercial voice cloning tools have since become widely available.
- Voice authentication can be bypassed. University of Waterloo researchers bypassed voice authentication systems with up to a 99% success rate after six attempts (IEEE Symposium on Security and Privacy, 2023; University of Waterloo). In April 2023, a Wall Street Journal columnist reported that an AI clone of her voice got past her bank’s voice check.
- Contact centers are a target. Pindrop recorded a 173% increase in synthetic voice calls between the first and fourth quarters of 2024 (Pindrop).
For insurers, the exposure is phone-based claim reporting, policy changes, payment redirection and recorded statements.
Synthetic Documents and Identities
- Onfido recorded a fivefold increase in digitally forged identities in 2023 compared with 2022 (Onfido, via Entrust).
- 58% of businesses in Regula’s 2024 survey had been deceived by fake or modified documents (Regula).
- A 2019 Federal Reserve white paper described synthetic identity fraud as a fast-growing problem and cited a widely reported estimate of US$6 billion in US lender losses in 2016.
- FinCEN’s November 2024 alert specifically flagged altered or synthetic identity documents made with generative AI (see Regulation below).
In insurance, the documents at risk include medical records, repair estimates and invoices, police reports, receipts and identity documents.
Can People Tell? Human Detection Research
| Study | Media | Finding |
|---|---|---|
| Nightingale & Farid, PNAS, 2022 | AI-generated faces | Participants scored 48.2%, close to chance, and 59% after training. Synthetic faces were rated more trustworthy than real ones. |
| Mai et al., UCL, PLOS ONE, 2023 | Synthetic speech | Listeners identified fake speech 73% of the time, in English and Mandarin alike; training helped only slightly (PLOS ONE). |
| University of Florida, 2024 | Audio deepfakes, 1,200 participants | Participants reached about 73% accuracy: better than chance, but often fooled by details such as accents and background noise (UF News). |
For claims teams, the takeaway is straightforward: visual or aural inspection alone gets at least one judgment in four wrong, even in controlled studies.
Detection Research
- Detectors can be attacked directly. Carlini and Farid showed that a synthetic-image detector with an AUC of 0.95 could be pushed as low as 0.0005 with imperceptible white-box perturbations, and to 0.22 with a black-box attack (Carlini & Farid, 2020). Layered detection and human review matter for this reason.
- Benchmark numbers do not transfer automatically. Detector performance depends on the media, compression and generation methods in the test set. Evaluate any detector on media like your own; see the lab-to-production gap and deepfake detection KPIs.
Regulation
- United States: FinCEN Alert FIN-2024-Alert004 (November 13, 2024) warned financial institutions about fraud schemes using deepfake media and listed red flags. The NAIC’s Model Bulletin on insurers’ use of AI had been adopted by 25 states and the District of Columbia as of August 31, 2026 (NAIC).
- European Union: EU AI Act Article 50 transparency duties, including disclosure of deepfakes, apply from August 2, 2026 (Mayer Brown summary).
- Australia: APRA’s CPS 230 Operational Risk Management standard has applied since July 1, 2025. The Insurance Council of Australia runs a dedicated counter-fraud function, the Insurance Crime Intelligence Network of Australia (ICINA).
What We Left Out
Earlier versions of this page carried several precise figures (market sizes, adoption percentages, per-incident costs and multi-year projections) that we could not trace to a published source. We removed them. If a statistic about deepfake fraud in insurance does not name a publisher, a year and a method, treat it as an estimate.
The trend in the verified data is consistent: deepfake attempts are rising fast from a small base, voice attacks on insurers are growing faster than in other sectors, and people cannot reliably spot synthetic media by eye or ear.
For corrections or newer releases, contact us.
Related Reading
- Insurance Fraud Statistics 2026: total fraud costs by country and line
- Insurance Deepfake Fraud Incidents: A Timeline: the cases and alerts behind these numbers
- The True Cost of Deepfake Fraud in Insurance: turning these figures into an exposure estimate
deetech™ is built around claims workflows: it analyzes images, video, audio and documents for signs of synthetic or manipulated media and gives claims and SIU teams reviewable findings. Request a demo.