Deepfake statistics 2026
Every figure on this page is computed from our incident database — 3,934 documented, source-cited deepfake incidents, updated daily. Nothing here is estimated or borrowed from a vendor report. Free to cite.
Documented incidents by year
Years before 2024 are excluded: the handful of landmark cases we hold from those years are a curated selection, not a representative sample.
By media type
| Type | Count | Share |
|---|---|---|
| Image / synthetic photo | 1,245 | 31.6% |
| Multi-modal | 1,082 | 27.5% |
| Video / face-swap | 1,044 | 26.5% |
| Voice / audio | 563 | 14.3% |
By attack pattern
Patterns overlap — one incident can be both election-related and a voice clone — so these do not sum to 100%.
| Pattern | Count | Share |
|---|---|---|
| Election deepfakes | 299 | 7.6% |
| Fabricated news footage | 297 | 7.5% |
| Celebrity deepfakes | 213 | 5.4% |
| Voice-clone scams | 164 | 4.2% |
| Deepfake CEO fraud | 79 | 2.0% |
| Investment & crypto deepfakes | 61 | 1.6% |
2026 month by month
Peak month so far: May 2026, with 410 documented incidents.
The final month is partial — data runs to 12 September 2026.
Methodology
An incident enters the database when a deepfake or AI-generated media event is documented by at least one citable public source — a news report, court filing, platform disclosure, or corroborated first-hand account. Entries are deduplicated by event, so a single case covered by twenty outlets counts once.
Incidents are dated by when the event occurred, not when we recorded it. Media type is assigned from the synthetic artefact itself; an incident involving both a cloned voice and a fabricated video is counted as multi-modal.
Read these as floors, not totals. Deepfake attacks are systematically under-reported — voice fraud against individuals and non-consensual imagery especially so. Coverage also deepens over time, which inflates recent years relative to older ones. Year-over-year growth here reflects both a real increase in attacks and an increase in what gets reported.
A subset of entries — those involving minors, non-consensual sexual imagery, or private individuals — is counted in these statistics but not published as browsable pages.
For the narrative behind these numbers — the fraud economy, the collapse in default digital trust, and the 2026 regulatory pivot — see the Deepfake Threat Report 2026.
Common questions
- How many deepfake incidents happened in 2026?
- Our database holds documented, source-cited incidents for 2026 and is updated daily. The live figures on this page are computed directly from that corpus, so they change as new incidents are verified. Because the year is still in progress, we publish both the year-to-date count and an annualised projection rather than a single headline number.
- Where does this data come from?
- Every incident is drawn from public reporting — news outlets, court filings, platform disclosures, and corroborated community reports — and each entry links its primary sources. Incidents are deduplicated so one event counted across many outlets appears once.
- Can I cite these statistics?
- Yes, freely. A suggested citation is at the bottom of this page. We keep URLs stable and recompute the figures daily, so please include the access date.
- What are the limits of this data?
- It counts documented incidents, not total incidents. Deepfake attacks are systematically under-reported — particularly voice fraud against individuals and non-consensual imagery — so treat every figure as a floor rather than a true total. Coverage also deepens over time, which inflates recent years relative to older ones.
Journalists and researchers: use this freely.
Published under CC BY 4.0. Figures are recomputed daily from the underlying database, so please include your access date. Corrections and additions are welcome.
https://www.detectdeepfakes.com/research/deepfake-statistics-2026,
accessed Sep 2026.