Deepfakes have appeared in 39% of the world’s elections held since fall 2023, with 133 incidents logged across 30 countries (German Marshall Fund, Spitting Images 2024-2026). The scale is wider still by other counts: 38 countries have faced election deepfakes since 2021, touching a combined population of 3.8 billion (Surfshark, 2025). Voter anxiety has kept pace, with 57% of US adults extremely or very concerned that AI would be used to fabricate election content (Pew Research Center, 2024). Yet the most cited enforcement action to date, the $6 million FCC fine over a cloned-voice Biden robocall, sits alongside academic findings that no 2024 vote showed measurable deepfake impact on results (Alan Turing Institute, 2024). This analysis consolidates data from the German Marshall Fund, Pew Research Center, Surfshark, and 13 other primary sources into the election-deepfake numbers that matter for 2026.
TL;DR
- 133 election deepfakes tracked across 30 countries since fall 2023 (German Marshall Fund, 2024-2026).
- 39% of elections held since fall 2023 featured a significant deepfake campaign (German Marshall Fund, 2024-2026).
- 38 countries have faced election deepfakes since 2021, affecting 3.8 billion people (Surfshark, 2025).
- 482 politically motivated deepfake incidents recorded in Q3 2025 alone (Resemble AI, 2025).
- Audio is the most common election-deepfake format, appearing in 69% of tracked instances (German Marshall Fund, 2024-2026).
- 57% of US adults are extremely or very concerned AI will spread fake election information (Pew Research Center, 2024).
- 58% of people worldwide worry about telling real from fake news online (Reuters Institute, 2025).
- FCC fine for the deepfake Biden robocall in New Hampshire: $6 million (FCC, 2024).
- Confirmed viral AI-disinformation cases in the 2024 UK general election: 16 (Alan Turing Institute, 2024).
- Deepfakes made up 5.86% of election images in a study of the 2025 Canadian election (ACM Web Conference, 2026).
- US states with laws regulating election deepfakes as of May 2026: 30 (Public Citizen, 2026).
- Cheap fakes were used seven times more often than AI deepfakes in 2024 elections (Knight First Amendment Institute, 2024).
1. How Many Election Deepfakes Have Actually Been Tracked
The headline counts are smaller than the alarm suggests, but they are rising and they are global. Named trackers with public methodology converge on a picture of steady, worldwide spread rather than a single flood. The German Marshall Fund has logged 133 election deepfakes across 30 countries, and found 39% of elections since fall 2023 carried a significant deepfake campaign (German Marshall Fund, 2024-2026). Resemble AI’s quarterly count adds granularity, recording 482 politically motivated incidents in Q3 2025 out of 2,031 total verified incidents, a 317% jump from the prior quarter. Crucially, when the Knight First Amendment Institute reviewed the 78 cases in the WIRED AI Elections Project, it found no deceptive intent in 39 of them, roughly half. For the broader synthetic-media picture, see our deepfake statistics roundup.
| Metric | Value | Source |
|---|---|---|
| Election deepfakes tracked across 30 countries (since fall 2023) | 133 | German Marshall Fund, 2024-2026 |
| Elections featuring a significant deepfake campaign | 39% | German Marshall Fund, 2024-2026 |
| Countries hit by election deepfakes since 2021 (3.8B people) | 38 | Surfshark, 2025 |
| Countries with elections since 2023 that saw deepfakes | 33 of 87 | Surfshark, 2025 |
| Politically motivated deepfake incidents, Q3 2025 | 482 | Resemble AI, 2025 |
| AI-use instances tracked across 2024 global elections | 78 | Knight First Amendment Institute, 2024 |
| 2024 tracked cases with no deceptive intent | 39 of 78 | Knight First Amendment Institute, 2024 |
| Times cheap fakes outnumbered AI deepfakes in 2024 | 7x | Knight First Amendment Institute, 2024 |
Context note: the Purdue-based Political Deepfakes Incidents Database catalogued 114 politically salient incidents dating back to 2017, showing synthetic political media predates the current generative-AI wave (AAAI, 2024).
2. Format and Spread: How Election Deepfakes Travel
The format mix upends a common assumption that deepfakes are mainly slick fake videos. In the German Marshall Fund’s dataset, audio is the single most common format, present in 69% of tracked instances, ahead of video at 55% and image at 20% (German Marshall Fund, 2024-2026). Audio is cheap to produce, hard to visually debunk, and travels well through voice channels. Distribution skews overwhelmingly to open social platforms, with private messaging apps a notable second vector, which complicates takedown and measurement.
| Metric | Value | Source |
|---|---|---|
| Election deepfakes using audio | 69% | German Marshall Fund, 2024-2026 |
| Election deepfakes using video | 55% | German Marshall Fund, 2024-2026 |
| Election deepfakes using image | 20% | German Marshall Fund, 2024-2026 |
| Incidents spread via social media | 92% | Surfshark, 2025 |
| Incidents spread via messaging apps | 32% | Surfshark, 2025 |
| Incidents spread via online targeted ads | 32% | Surfshark, 2025 |
| Platform share: X / Facebook / WhatsApp | 53% / 39% / 29% | Surfshark, 2025 |
| Estimated AI-content spend, India 2024 election | $50 million | Surfshark, 2025 |
Outlier note: phone calls and robocalls accounted for just 5% of distribution channels (Surfshark, 2025), a small share that nonetheless produced the most consequential US enforcement case to date.
3. Voter Trust and Concern About AI Election Content
Public concern runs far ahead of documented harm, and that gap is itself a finding. 57% of US adults told Pew they were extremely or very concerned AI would be used to create and distribute fake election information, with almost no partisan split: 56% of Republicans and 58% of Democrats (Pew Research Center, 2024). Only 20% expressed confidence that major platforms would prevent misuse. Globally, the Reuters Institute found 58% worried about distinguishing real from fake news online, and respondents rated influencers and politicians as equal top sources of false information.
| Metric | Value | Source |
|---|---|---|
| US adults very concerned AI will fake election info | 57% | Pew Research Center, 2024 |
| Say AI will be used mostly for bad purposes / mostly good | 39% / 5% | Pew Research Center, 2024 |
| Confident tech firms will prevent platform misuse | 20% | Pew Research Center, 2024 |
| Worried about telling real from fake news online | 58% | Reuters Institute, 2025 |
| Rate influencers / politicians as top false-info threat | 47% / 47% | Reuters Institute, 2025 |
| Californians very confident telling real from fake content | 8% | Carnegie Endowment, 2026 |
| Californians very concerned about deepfakes in elections | 57% | Carnegie Endowment, 2026 |
| Mis/disinformation ranking among short-term global risks | No. 1, 2nd year | World Economic Forum, 2025 |
Context note: the Pew survey polled 9,720 US adults between August 26 and September 2, 2024, and the World Economic Forum has now ranked misinformation and disinformation the top short-term global risk two years running.
4. Voice Deepfakes and Election Robocalls
Voice is where synthetic election media has drawn its sharpest regulatory response. The FCC issued a $6 million fine against political operative Steve Kramer over January 2024 robocalls that used a cloned voice of President Biden urging New Hampshire voters to skip the primary, a call that reached up to 25,000 people (FCC, 2024). The carrier that transmitted the calls, Lingo Telecom, agreed to a separate $1 million civil penalty. A New Hampshire state jury later acquitted Kramer of criminal voter-suppression and impersonation charges, underscoring how far the legal picture remains unsettled. For the wider phone-fraud context, see our robocall statistics and vishing statistics.
| Metric | Value | Source |
|---|---|---|
| FCC fine against Steve Kramer (NH robocall) | $6 million | FCC, 2024 |
| New Hampshire voters targeted by the deepfake robocall | up to 25,000 | FCC, 2024 |
| Lingo Telecom civil penalty | $1 million | FCC, 2024 |
| Audio share of tracked election deepfakes | 69% | German Marshall Fund, 2024-2026 |
| Robocalls share of deepfake distribution channels | 5% | Surfshark, 2025 |
| Kramer state criminal-charge jury verdict (June 2025) | acquitted | NHPR, 2025 |
Context note: the FCC acted within weeks of the calls and its subsequent ruling clarified that AI-generated voices in robocalls fall under existing telemarketing law, the first federal action of its kind tied to an election deepfake.
5. What the Evidence Says About Real-World Impact
Documented electoral effect has been hard to find, even as incident counts climb. The Alan Turing Institute identified just 16 confirmed viral AI-disinformation cases in the 2024 UK general election and 11 across the EU and French elections combined, with no evidence any of them changed a result (Alan Turing Institute, 2024). The most granular academic measurement, a peer-reviewed study of the 2025 Canadian federal election, found deepfakes made up 5.86% of election images yet drew only 0.52% of views on X, and nearly half were benign or non-political. Detection remains the constraint on measuring the rest; see our deepfake detection statistics.
| Metric | Value | Source |
|---|---|---|
| Confirmed viral AI-disinfo cases, 2024 UK general election | 16 | Alan Turing Institute, 2024 |
| Confirmed viral AI-disinfo cases, EU + French elections | 11 | Alan Turing Institute, 2024 |
| Measurable impact on UK/French/EU results | none found | Alan Turing Institute, 2024 |
| Election images that were deepfakes, Canada 2025 (of 187,778 posts) | 5.86% | ACM Web Conference, 2026 |
| Deepfake rate on X vs Bluesky, Canada 2025 | 7.9% vs 2.2% | ACM Web Conference, 2026 |
| Synthetic-image sharing: right-leaning vs left-leaning accounts | 9.24% vs 3.87% | ACM Web Conference, 2026 |
| Deepfakes as share of all views on X, Canada 2025 | 0.52% | ACM Web Conference, 2026 |
| Detected deepfakes that were benign or non-political | 49.41% | ACM Web Conference, 2026 |
Context note: the clearest exception is Romania, where the Constitutional Court annulled the first round of its 2024 presidential election on December 6, 2024, citing coordinated inauthentic amplification and AI-generated content, the first such annulment in the EU (OECD.AI incident database, 2024).
6. Election Deepfake Laws and Regulation
Lawmaking has accelerated sharply, mostly through disclosure rules rather than outright bans, reflecting free-speech constraints. As of May 2026, 30 US states had enacted laws addressing deepfakes in political communications, up from 28 at the start of the year, with Maryland becoming the 30th (Public Citizen, 2026). The pace was faster still in 2025, when the count of states regulating political deepfakes climbed from 21 to 28 in six months (R Street Institute, 2025). Federal and EU transparency requirements are arriving in parallel.
| Metric | Value | Source |
|---|---|---|
| US states with election-deepfake laws (May 2026) | 30 | Public Citizen, 2026 |
| US states with such laws at start of 2026 | 28 | Public Citizen, 2026 |
| 30th state to enact protections | Maryland | Public Citizen, 2026 |
| States regulating political deepfakes, Jan to July 2025 | 21 to 28 | R Street Institute, 2025 |
| EU AI Act Article 50 transparency rules apply from | 2 August 2026 | EU AI Act, Article 50 |
| California AI Transparency Act (AB 853) signed | 13 October 2025 | Carnegie Endowment, 2026 |
| Pennsylvania Act 35 deepfake law signed | 7 July 2025 | Resemble AI, 2025 |
| Washington State forged-digital-likeness law effective | 27 July 2025 | Resemble AI, 2025 |
Context note: Public Citizen reports these measures passed with bipartisan support in every state that enacted them, and most rely on disclaimers on AI-generated political content rather than prohibitions.
Summary: Deepfakes In Elections by the Numbers
| Metric | Value | Source |
|---|---|---|
| Election deepfakes tracked across 30 countries | 133 | German Marshall Fund, 2024-2026 |
| Elections with a significant deepfake campaign | 39% | German Marshall Fund, 2024-2026 |
| Countries hit by election deepfakes since 2021 | 38 | Surfshark, 2025 |
| Population in affected countries | 3.8 billion | Surfshark, 2025 |
| Politically motivated deepfake incidents, Q3 2025 | 482 | Resemble AI, 2025 |
| Audio share of election deepfakes | 69% | German Marshall Fund, 2024-2026 |
| Incidents spread via social media | 92% | Surfshark, 2025 |
| US adults very concerned about AI election fakes | 57% | Pew Research Center, 2024 |
| Confident tech firms prevent platform misuse | 20% | Pew Research Center, 2024 |
| Worried about real vs fake news online (global) | 58% | Reuters Institute, 2025 |
| Californians very confident telling real from fake | 8% | Carnegie Endowment, 2026 |
| FCC fine for NH deepfake Biden robocall | $6 million | FCC, 2024 |
| Voters targeted by that robocall | up to 25,000 | FCC, 2024 |
| Confirmed viral AI-disinfo cases, 2024 UK election | 16 | Alan Turing Institute, 2024 |
| Election images that were deepfakes, Canada 2025 | 5.86% | ACM Web Conference, 2026 |
| Cheap fakes vs AI deepfakes usage, 2024 | 7x more | Knight First Amendment Institute, 2024 |
| US states with election-deepfake laws (May 2026) | 30 | Public Citizen, 2026 |
| EU AI Act transparency rules apply from | 2 August 2026 | EU AI Act, Article 50 |
Methodology and Sources
Data was gathered by prioritizing primary sources published in 2024-2026, tracing every figure back to the original tracker, survey, filing, court record, or peer-reviewed study rather than secondary aggregators. Sources cited:
- German Marshall Fund, Spitting Images: Tracking Deepfakes and Generative AI in Elections (topline data, 2024-2026)
- Surfshark Research, Election-Related Deepfakes (2025)
- Resemble AI, Q3 2025 Deepfake Incident Report (2025)
- Knight First Amendment Institute, We Looked at 78 Election Deepfakes (2024)
- Political Deepfakes Incidents Database (Walker, Schiff and Schiff, Purdue University / AAAI ICWSM, 2024)
- Pew Research Center, Concern Over the Impact of AI on the 2024 Presidential Campaign (2024)
- Reuters Institute for the Study of Journalism, Digital News Report 2025
- Carnegie Endowment for International Peace, AI and Democracy: Mapping the Intersections (2026)
- World Economic Forum, Global Risks Report 2025
- Federal Communications Commission, FCC Issues $6M Fine for N.H. Robocalls (2024)
- New Hampshire Public Radio, Kramer acquitted on robocall charges (2025)
- Alan Turing Institute, No evidence AI disinformation or deepfakes impacted UK, French or European elections (2024)
- ACM Web Conference 2026, Deepfakes in the 2025 Canadian Election: Prevalence, Partisanship, and Platform Dynamics
- Public Citizen, 30 States Now Have Laws to Regulate Election Deepfakes (2026)
- R Street Institute, Update on 2025 State Legislation to Regulate Election Deepfakes (2025)
- OECD.AI Incidents Monitor, AI-driven social media interference annuls Romania presidential vote (2024)
- EU AI Act, Article 50 (transparency obligations for AI-generated content)
Data watch: several of these sources refresh on predictable cycles. The German Marshall Fund’s Spitting Images tracker and Public Citizen’s state-law tracker update continuously; Resemble AI publishes quarterly incident reports (Q4 2025 expected next); the Reuters Institute Digital News Report and the World Economic Forum Global Risks Report both publish annually in mid-January and June; Pew Research Center runs recurring election-AI surveys around major US votes, with fresh 2026 midterm data likely. Expect the 30-state legislation count to keep moving ahead of the November 2026 US midterms.
Last updated: July 10, 2026.
We review and update this page quarterly as new data is published.