Fake videos using the voice and image of well-known figures have exploded on social media in recent weeks, especially with the arrival of elections. Recent cases show the power of artificial intelligence (AI) in this movement – it can invent political statements, alter audio, and create scenes so realistic they easily deceive someone scrolling on their phone screen.
This tactic exploits the influence of popular figures to lend credibility to invented content. In this way, high-profile names begin to integrate manipulations that combine real and false elements to deceive the public.
Advanced deepfakes manipulate faces and voices with high perfection, making it very difficult for voters to verify false information. – Credit: Image generated by AIAI in service of deceptive political discourse
One of the examples involves Caetano Veloso and Neymar Jr. In the case of the singer, a recording circulating on TikTok was created to appear as if he said 'Lula saved this country from hatred' and supported the continuation of the government in 2026. It's a deceptive montage, as the singer never made that statement.
The forgery exploited true elements to gain an appearance of authenticity. Caetano actually participated in 2026 in a Lula campaign material where artists appear saying 'Bora, Lula!'. However, in the fake video, the singer's image was reused with a new audio, produced with AI assistance, containing words he never uttered. Caetano's own advisory team confirmed to DW that the content was false.
With Neymar, the strategy was similar, but with an opposite political bias. A manipulated video made the player claim that he was following a supposed investigation against Jair Bolsonaro and that Lula and the Supreme Federal Court (STF) Minister Alexandre de Moraes would be "afraid of the president's power." The statement was never made by the player. The video used the athlete's image to attribute to him a political statement that did not occur.
In this case, there was also a real element that helped give the forgery an appearance of authenticity. Neymar declared his vote for Jair Bolsonaro in the 2022 elections, and in 2026, appeared in a real photograph autographing a Brazilian national team jersey for Flávio Bolsonaro, the son of the former president and presidential candidate. However, the real image does not prove that the player made the political statement presented in the false video.
The combination of true facts and fabricated content is one of the aspects that make these cases particularly difficult to identify. Those who already know that Caetano participated in a Lula campaign or that Neymar had a public relationship with Jair Bolsonaro may consider the video "plausible" before even checking its origin. The mix complicates verification because the false content relies on facts that the public already knows are true.
An infographic shows how false content gains credibility by mixing real facts, AI manipulations, and invented elements. – Credit: AI-generated image
The use of images and voices of well-known people is not limited to political propaganda. Tony Ramos was also a victim of deepfake, as reported by Olhar Digital. In a recent appearance on the TV Cultura program Roda Viva, the actor revealed that his image had been used in a fake video alongside Ana Maria Braga to sell a medication. According to the report, his speech was dubbed and used to give an appearance of authenticity to an advertisement he had not made.A false image of Alexandre de Moraes and Edson Fachin, also a minister of the STF, also circulated in recent days. The publication suggested they were discussing, but the image was generated by AI, according to fact-checking platforms. The incident shows that manipulation does not necessarily depend on video: a photograph can also create a non-existent situation and produce a misleading narrative.Resolution 23.610/2019 of the Superior Electoral Court (TSE) determines that synthetic content produced or manipulated by AI must be identified when used in propaganda. The regulation also prohibits, in electoral propaganda, deepfakes that create, replace, or alter the image or voice of people to favor or harm candidacies. At the beginning of September, the TSE also established criteria to define this type of content.According to the agency, deepfake is synthetic content generated or altered by artificial intelligence or equivalent technology, with a high degree of realism, capable of creating, reproducing, or modifying the image, voice, or statement of any person – alive, deceased, or fictional. Under the established understanding, the explicit prohibition of the tool applies to content configured as election propaganda.
AI did not eliminate old manipulation techniques. Real videos continue to be taken out of context, cut, or combined with other audios to create situations that never happened. The UOL Confere check published on this Wednesday (30) revealed manipulation in an old video showing a man attacking a priest in Rio de Janeiro, in 2022.
In the original recording, the aggression occurred after a personal disagreement; in the altered version that circulated this week, the original audio was replaced by a speech about national sovereignty. The editing simulated that the religious figure had been attacked for political motives – a contextually removed clip that accumulated over 100,000 views on Facebook until the article's publication.
Simple precautions to recognize manipulated content, avoid scams, and interrupt the spread of false information on social media. – Credit: AI-generated image
Rise in deepfakes fuels disinformation and scams in Brazil
Data from the First Disinformation Panorama in Brazil, published by Agência Lupa in February, shows that false content generated or manipulated with AI identified in the organization's checks went from 39 in 2024 to 159 in 2025 – a 308% increase. And a recently released study by Ekō, an international corporate responsibility organization, identified at least R$1.1 million paid by 26 profiles to boost election advertisements that, according to the research, violated Meta's rules and Brazilian legislation, including deepfakes without identification.
The same combination of manipulated image, political content, and appearance of news can also be used for something beyond disinformation: applying financial fraud. Recently, images generated by AI and the simulation of a news portal suggested that businessman Luciano Hang, owner of Havan stores, had suffered a "attack" at the order of former Minister of Finance Fernando Haddad, current PT candidate for São Paulo governor.
The story appealed to political rivalry to direct the public to a fake website with the promise of profits up to R$10,000 per day, where users were induced to make deposits or provide personal data.
With artificial intelligence, criminals can manipulate faces and videos, creating fake content that can easily pass as authentic on social networks. – Credit: Tero Vesalainen/ShutterstockWilder Martins, a cybersecurity specialist at EXA, developer of solutions against virtual threats, emphasizes that current technology perfectly simulates reality, creates fake links and websites, very similar to the originals. “Therefore, the main defense barrier has changed: the question should no longer just be ‘does this seem true?’ but rather ‘why are they asking me to click a link, provide sensitive data, or make a transfer?’”.
How to identify manipulated content and avoid traps
According to Martins, the same structure can be adapted to different topics that gain relevance during the electoral period. The greater circulation of political information creates new opportunities for criminals to exploit candidate names, authorities, institutions, and media outlets. Among the possibilities are messages about voter titles, biometrics, polling locations, election polls, fake fines, campaign fundraising, and manipulated content using the voice or image of public figures.
“If the message creates an emergency (whether to avoid a block, ensure a right, or demand payment), stop immediately,” warns the expert. “This is the classic trigger of fraud. Your main defense is not following the shortcut offered by the fraudster. Close the message and access verification tools and software, both paid and free, such as EVA [artificial intelligence from EXA that analyzes suspicious WhatsApp content] to confirm the information before making any transaction”.
With the advancement of AI, identifying fraud solely through visual or auditory cues has become more difficult. Failures that previously revealed some manipulations are less evident, and verification must go beyond the appearance of the content. Instead of looking for only visual flaws, the guidance is to check the content's trail: confirming if the person actually made that statement in official channels, verifying if media outlets reported the fact, and being suspicious of controversial videos that circulate without a clear origin.
Flávia Correia is a journalist at Olhar Digital, covering Science and Space.
