By: Andre Nicola
On October 4, 155 million Brazilians will head to the polls in what will be the country's first major election shaped by widespread generative AI — deepfakes, synthetic audio and visual content, and chatbots capable of directly engaging voters. The presidential race is expected to be a tense contest between incumbent President Luiz Inácio Lula da Silva and Senator Flavio Nantes Bolsonaro, son of former president Jair Bolsonaro, unfolding in a political climate defined by deep polarization, institutional distrust, and years of escalating digital disinformation. Unlike previous elections where the primary concern was social media manipulation, 2026 will test how Brazil, and other large democracies, respond to this new generative AI challenge.
Brazil is uniquely positioned to confront this challenge. Following the 2018 elections, where the political power of WhatsApp misinformation networks was laid bare, Brazil's Superior Electoral Court (TSE), which holds unusually sweeping powers compared to electoral bodies elsewhere, emerged as one of the world's most assertive electoral management bodies. After former president Bolsonaro systematically attacked the integrity of the electoral system, the TSE ruled him ineligible for public office and has since treated digital disinformation as a threat to democratic stability rather than a platform moderation problem.
These experiences shaped the TSE's AI electoral framework, arguably one of the most detailed anywhere in the world. Approved in 2024 and updated March 2026, the rules ban deceptive deepfakes, require labeling of AI-generated campaign content, prohibit AI tools from recommending specific candidates, impose a blackout on synthetic content in the days surrounding the election, and establish platform liability for non-compliance. Candidates who deploy deepfakes to manipulate voters risk having their registration annulled. What makes this framework especially remarkable is not only its scope, but the fact that it was built without a comprehensive national AI law. The TSE used its constitutional authority over elections to construct targeted safeguards rather than waiting for Congress to act. This stands in contrast to the European Union, whose AI Act, despite its ambition, contains remarkably little specific to electoral integrity, and to the United States, where no equivalent federal electoral AI framework exists at all.
One feature is particularly notable. Brazil's rules shift the burden of proof onto campaigns and platforms when suspected synthetic content appears online. In practice, if a campaign video is flagged as potentially synthetic, it is the campaign's responsibility to prove it is not, rather than regulators having to prove that it is. In theory, this allows regulators to act before manipulated material spreads. The rationale is simple: as AI advances and viral synthetic content becomes more prevalent, democratic institutions cannot operate at the sluggish pace of traditional judicial proceedings.
Yet significant weaknesses remain. The framework concentrates accountability on campaigns and platforms while largely exempting the companies that build AI systems, despite the fact that their products directly shape electoral outcomes. In practice, enforcement has already proven problematic. Weeks after the rules came into effect, fact-checking conducted by Agence France-Presse (AFP) found that ChatGPT, Grok, and Gemini were still ranking and recommending candidates when prompted. When flagged content is removed, it can reappear immediately in a new form. But perhaps most structurally difficult is the challenge of encrypted messaging. In Brazil, as across much of Latin America, WhatsApp and Telegram are the primary channels for political communication and because disinformation circulates through trusted networks of family, church, and community groups rather than public feeds, it is nearly impossible to monitor or moderate systematically.
These weaknesses are compounded by Brazil's legislative vacuum. PL 2338, the national AI bill, passed the Senate in December 2024 but is still under review by a special committee in the Chamber of Deputies, with no expected timeline. The delay is not accidental and has been resisted by U.S.-based and multinational big tech corporations. The court can mobilize quickly, but it lacks the permanence, jurisdiction, and comprehensive oversight functions that only national legislation can provide, legislation that the industry most affected is actively working to delay.
Brazil's experiment nonetheless offers important lessons. Governments do not need to wait for comprehensive AI legislation before acting on electoral integrity, and given the pace of technological change, waiting may prove both politically unrealistic and dangerous. Brazil is not alone in recognizing this; the Philippines' COMELEC adopted similarly targeted electoral AI rules for its 2025 elections without a national AI law, demonstrating that electoral bodies can move faster than legislatures when the stakes are clear. Yet both cases also reveal the same structural ceiling — administrative rules without legislative backing struggle to hold developers accountable, reach encrypted channels, or impose meaningful deterrence. The United States, heading into its midterm cycle with no federal AI law and a fragmented patchwork of state-level rules, has yet to reckon with this question at all.
The question is no longer whether AI will shape elections. It is whether governments will build the right safeguards before, rather than after, elections can be manipulated.
Andre Nicola is a Program Coordinator for Latin America at ORF America.