digital democracyartificial intelligence

Election-related deepfakes affected over 4.2B people globally

Election deepfakes have gone from rare incidents to a common threat to democratic processes worldwide. As AI tools have become cheaper and easier to access, deepfakes have spread rapidly across multiple countries and platforms. They're now a regular feature of elections globally, used to mislead voters and undermine trust in the electoral process. Our research reveals how widespread this problem has become, which regions are most affected, and where deepfakes are spreading the fastest.

Key insights

• Our research identified 235 election-related deepfake incidents reported by news media across 46 countries with a combined population of 4.2 billion. The earliest incidents in our dataset appeared in India and the United States in 2020, where manipulated videos circulated during elections and targeted voters. Activity remained limited in 2021–2022, with nine incidents across eight countries, before accelerating sharply in 2023 as inexpensive, accessible AI tools became more widely available. From 2024 onward, we have recorded 52–74 incidents per year, suggesting that deepfakes have evolved from isolated cases into a persistent feature of elections worldwide.

• The United States recorded the highest number of election-related deepfake incidents in our dataset, with 87 cases, accounting for 37% of the global total. This is nearly nine times as many as the United Kingdom, the most affected European country, with 10 incidents. US election-related deepfake activity peaked at 34 incidents between January and September 2026, amid primary elections and preparations for the November midterm elections. Despite covering only nine months, the 2026 total was already 21% higher than the 28 incidents recorded throughout 2024, when the country held its presidential election.

• Europe accounted for 38 election-related deepfake incidents in our dataset, affecting 16 of the region’s 44 countries. The United Kingdom recorded the highest number, with 10 incidents, followed by Slovakia with five and Kosovo with four. Ireland and France each recorded three incidents, while Poland and Moldova each recorded two. Nine other countries recorded one incident each: Switzerland, Spain, Slovenia, Romania, the Netherlands, Malta, Germany, Croatia, and Bosnia and Herzegovina. Overall, election-related deepfakes affected more than one-third of European countries, but half of the region’s incidents were concentrated in just three countries.

• The Asia-Pacific region accounted for 73 election-related deepfake incidents across 15 countries. India recorded the highest number, with 19 incidents, followed by South Korea with 16, Taiwan with 13, and Indonesia with seven. Together, these four countries accounted for 80% of all incidents recorded in Asia, showing that the region’s deepfake activity was heavily concentrated in a small number of countries. Oceania recorded only four incidents, with three in Australia and one in New Zealand.

• Social media was the dominant distribution channel for election-related deepfakes, accounting for 84% of incidents in which a channel could be identified. News and broadcast media ranked second at 11%, while messaging apps, phone calls, and websites accounted for the remainder. Among incidents for which a specific social media platform was identified, X, formerly Twitter, was mentioned most frequently, appearing in 49% of cases. TikTok followed at 29%, Facebook at 22%, Instagram at 12%, and YouTube at 7%.

Methodology and sources

This study used data from the AI Incident Database, Resemble AI, and the OECD to create a combined dataset covering deepfake incidents from January 2020 to September 2026. Incidents were included if they involved synthetic video, images, or audio, were documented in news media reports, and were related to elections.

Each incident was classified by target country, origin, and distribution platform. The findings represent a conservative estimate based on publicly reported incidents, while the source databases included records drawn from media coverage in multiple languages.

We also included relevant incidents identified in Surfshark’s 2025 election deepfake research¹, particularly those involving countries not represented in the three source databases. However, for the 2026 edition, we revamped the data collection process and did not rely on the methodology used in the 2025 study.

Data was collected from:

References: