AI-generated political deepfakes spread widely in Brazil ahead of October 2026 election
A study by Agência Lupa and FAAP researchers found that at least 40 AI-generated videos about Brazilian politics went viral between February and June 2026, averaging two per week, with President Lula appearing in 17 deepfakes that included false claims about cutting pensions and imprisoning opponents. Detection tools proved unreliable, with researchers unable to confirm or rule out AI creation in 125 additional pieces of content.
What happened
Between February and June 2026, researchers from Agência Lupa and students from Fundação Armando Alvares Penteado (FAAP) monitored social media feeds on Instagram, Facebook, X, TikTok and WhatsApp and collected 226 pieces of content suspected of being created or manipulated by AI. Of these, 101 were confirmed to be AI-generated, with approximately 40% relating to politics.
At least 40 AI-generated videos about the political-electoral scenario went viral during the monitoring period, averaging two per week. The materials included deepfakes of famous people declaring support or asking for votes for candidates, false electoral surveys presented by AI avatars, and humorous videos with images of politicians.
Reported impact
President Lula was the most frequently depicted Brazilian politician in the synthetic content, appearing in 17 of the AI-generated materials identified, including false videos in which he supposedly promised to cut pensions and imprison opponents. Former president Jair Bolsonaro appeared in eight pieces, including content that used his image to promote senator Flávio Bolsonaro's presidential candidacy.
Researcher Beatriz Farrugia stated that the results show AI use is already part of Brazilians' informational routine, and noted growing concern about voters' ability to distinguish authentic from artificial content amid this volume of materials.
Index analysis
Detection tools showed inconsistent results, with the same video or image receiving different diagnoses about AI use in 20 cases tested. An additional 125 pieces of content remained inconclusive, as researchers could not confirm or rule out AI creation.
Farrugia characterized the rapid circulation of these materials without adequate oversight as 'a concerning fragility,' citing a mismatch between the speed at which AI generation tools advance and the ability to detect their output.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.