The Study’s Figures
AI-generated deepfakes linked to election races are growing rapidly and are being identified as a threat to election integrity. That is according to a study by cybersecurity company Surfshark, reported by ETNews: between January and September 2026, 66 cases of election deepfakes were reported, already 27% higher than the 52 recorded for all of 2025. The research covers the period from January 2020 to September 2026 and catalogued 235 cases reported by the press in 46 countries, with a combined population of about 4.2 billion people.
The United States is the country with the most cases, 87, accounting for 37% of the total. India follows with 19, and South Korea is third worldwide with 16 cases since 2021. The cited ranking is rounded out by Taiwan with 13 and Indonesia with 7. In the Asia-Pacific region, 73 cases were recorded across 15 countries, concentrated mainly in the four states with the highest number of episodes. The first election deepfakes date back to 2020 in India and the United States; in 2021–2022 there were only 9 cases in 8 countries, while growth became evident from 2023.
The South Korean Case and the Dissemination Channels
In South Korea, the first documented episode dates back to 2021, when a candidate used his own deepfake avatar for an election campaign. Subsequently, manipulated videos and images intended to mock or damage the reputation of political figures multiplied, spreading on social networks. According to the reported data, ahead of the 2025 presidential elections, cases rose to about 9,800, twenty times more than the previous year. The government introduced specific rules against the abuse of election deepfakes, with financial penalties, investigative activities and detection tools; nevertheless, during the June 2026 local elections, requests to remove deepfake content exceeded 10,000, a sign that the burden on the authorities remains high.
As for dissemination channels, 84% of cases where the modality is known circulated through social networks, 11% through news outlets and broadcasters, and the rest through messaging apps, phone calls and websites. Among the identified platforms, X (formerly Twitter) accounts for 49%, TikTok for 29%, Facebook for 22%, Instagram for 12% and YouTube for 7%; the percentages can exceed 100% because a case can circulate on multiple platforms. In the United States, cases between January and September 2026 totaled 34, the peak, 21% more than in the 2024 presidential election year. In Europe, the United Kingdom is the country with the most episodes, 10, compared with 67 in the United States in the period considered.
The Explanations and Limitations of the Research
Luis Costa, Surfshark’s head of research, attributes the spread to the democratization of production tools: whereas a studio was once needed, today an app and a few minutes are enough, and the same method also appears in local elections. Costa also notes that manipulated content aims to provoke strong emotions, such as anger and surprise, so that users share it before verifying. His advice is to check whether reliable news outlets have reported the suspicious content and, if not, to refrain from spreading it. The study combines the AI Incident Database, Resemble AI and OECD data and examines only press-confirmed cases that use synthetic video, images or audio. Surfshark specifies that this is a conservative estimate: content actually in circulation but not reported by the media could make the phenomenon broader than measured. The real scale of dissemination therefore remains to be verified, while the available data indicate that the ability to ascertain the authenticity of election information and to quickly block manipulated content has become a technical and regulatory priority.
