AI-generated content, such as fabricated voters and deepfakes, is significantly impacting the Brazilian electoral campaign, particularly on platforms like TikTok. A survey by Aos Fatos identified at least 50 pieces of content simulating interviews and testimonials about presidential candidates Lula (PT) and Flávio Bolsonaro (PL), accumulating 7.4 million views. These videos often feature AI-generated individuals praising candidates, associating Lula with social benefits and Bolsonaro with political renewal and anti-corruption efforts. A significant concern is that 18% of these videos lacked any disclosure of AI manipulation, violating platform rules.
The use of synthetic characters in campaigns has already reached the electoral courts. In April, the Workers' Party (PT) filed a complaint with the Superior Electoral Court (TSE) against "Dona Maria," an AI-created character that gained millions of views criticizing the party and the Lula government. The TSE's Resolution 23.610/2019 prohibits the electoral use of synthetic audio or video that creates or manipulates the image or voice of living, deceased, or fictional individuals, extending beyond traditional deepfakes. This means that even before the official campaign period, such practices can be deemed illegal.
The problem extends to deepfakes of the candidates themselves. The Estadão newspaper monitored profiles supporting Lula and Flávio Bolsonaro, identifying 25 deepfake videos that had millions of interactions. Of these, 22 did not indicate the use of AI technology, despite TSE rules requiring such disclosure. The TSE has stated that it does not proactively monitor all synthetic content but will address cases involving disinformation against electoral integrity. Minister André Mendonça of the TSE recently ordered the removal of a deepfake video associating Flávio Bolsonaro with a banking executive and has proposed new rules for the 2026 elections, distinguishing between general synthetic content (which is permitted with transparency) and deepfakes (which are strictly prohibited to favor or harm candidates).
The new regulations for 2026 emphasize transparency, requiring explicit, prominent, and accessible identification for all AI-generated content used in electoral propaganda. They also prohibit AI content using the image or voice of candidates or public figures 72 hours before and 24 hours after an election. Furthermore, AI providers cannot rank, recommend, or prioritize candidates or campaigns, nor can they issue opinions or indicate electoral preferences. A significant change is the inversion of the burden of proof, meaning those using AI in electoral contexts may need to verify the origin and veracity of their content if questioned. An observatory noted that 40% of 357 identified political AI posts contained misinformation, and 67% attacked politicians, with 65% lacking the required AI usage disclosure.
