AI in elections: 6 real cases show how deepfakes have already challenged democracies around the world

'Deepfake': man makes pornographic montages of young man who ignored his flirtations The debate about electoral disinformation is not new in Brazil. But the spread of tools capable of producing highly realistic content has raised the concerns of electoral authorities, researchers and digital platforms. In recent years, tools capable of producing synthetic videos, images and voices, so-called deepfakes, have gone from experimental use to the daily lives of millions of people. And with just a few days to go before the official start of the election campaign, there are still no definitive answers on how to ensure the safe and responsible use of generative artificial intelligence. On Thursday (13), the ministers of the Superior Electoral Court (TSE) postponed a meeting to try to reach a consensus on the punishment for the use of AI in elections. The TSE will decide whether the presidential campaign of Flávio Bolsonaro (PL) committed any irregularity by using a video of former president Jair Bolsonaro (PL) produced with artificial intelligence (AI) during the PL party convention that made his candidacy for the Palácio do Planalto official at the end of July. A new date for the meeting has not yet been announced. Sam Stockwell, a researcher at the Center for Emerging Technologies and Security (CETaS), at the Alan Turing Institute, in the United Kingdom, studied several cases in which artificial intelligence tools were used during election campaigns in recent years. In Ireland, a deepfake released just days before the 2025 presidential election showed the main candidate giving up the campaign. In the same year, other videos spread false information about security threats on the day of parliamentary elections in Germany. But according to him, it is not possible to point out a direct link between the use of AI tools and the result of an election, as there are always many factors involved in a race. “To date, there is no conclusive evidence that AI tools have profoundly altered the outcome of any election,” he says. For Stockwell, the real danger of AI is not in directly altering votes, but rather in the subtle and ongoing effect of sowing confusion about what is real, eroding trust in institutions and information, and inflaming political divisions and polarization. "AI is eroding trust in our information space and, crucially, also inflaming political divisions at a time when polarization and violence in the real world are, obviously, at very high levels," he said in an interview with BBC News Brasil. And recent experiences from other countries help to illustrate the challenges that may arise during the Brazilian electoral dispute. Check out six examples of how deepfakes were used to spread misinformation around the world. False withdrawal 3 days before the election One of the most cited examples when the subject of deepfakes in elections is the one that hit the current president of Ireland, Catherine Connolly, last year. When the montage was released, in October 2025, she was running for President as an independent candidate. Election polls pointed to Connolly as one of the favorites, but three days before the vote, a video in which she appeared to withdraw from the race began circulating on social media. The fake video imitated a broadcast from RTÉ News, Ireland's largest and most popular news source. In the images, an AI-created version of Connolly announces his withdrawal from the election, while supporters boo and lament the decision. Soon after the video began circulating, the candidate's campaign assured her that she was still in the race and alerted social media platforms to have it removed or properly identified as created by AI. Connolly called the video a "shameful attempt to mislead voters and undermine our democracy." Montage indicates that the candidate’s video was generated by AI. BBC/Facebook At the time, Alan Smeaton, emeritus professor of computing at Dublin City University and a member of the Irish government's AI Advisory Board, told BBC Verify that there were a few "tiny clues" that the video was fake. He pointed out details such as a slight overlapping of words, both in Catherine Connolly's speech and that of the journalists portrayed, as well as flaws in the background of the images. Despite the deepfake, Connoly won the elections, with 63% of the votes. "Although it wasn't enough to change the outcome, this is a very worrying example of deepfake, because there was a hijacking of a trusted national institution and its credibility to try to make this content seem much more convincing or persuasive," says Sam Stockwell, referring to broadcaster RTÉ News. Flood of videos in Ecuador Deepfakes simulating television channels were also common in the run-up to Ecuador's February 2025 presidential elections. Using fake logos, videos designed to imitate broadcasters such as CNN, France 24 and Deutsche Welle falsely implicated candidates in scandals and spread allegations of voter fraud. In one of them, a false presenter says that a corruption scandal "exploded in the face" of Daniel Noboa, who was running for re-election at the time. Another showed a presenter saying that Noboa had paid US$1 million to attend Donald Trump's inauguration in the United States. Another video showed an AI-created version of candidate Luisa González, who was running for the Citizen Revolution movement, saying that if she were elected she would implement a drastic tax increase. The contents were denied by news verification sites and by the campaign itself. González was also the target of another AI-generated video, in which he appeared praising Venezuelan President Nicolás Maduro. After expert analysis, the video was identified as a deepfake. Daniel Noboa was re-elected on April 13 last year, with around 56%