AI and Democracy in Brazil: Where to Draw the Line?
Elections, regardless of the time and place in which they occur, rely on a key pillar for candidates: their ability to campaign. While this may seem obvious, artificial intelligence has completely altered how campaigns are carried out and has attacked this foundational component of elections. A candidate can now appear digitally, in a vast array of formats, communicating a political message to voters without needing to physically participate in their own campaign. Brazilian voters have recently been subjected to this development, forcing its government to enforce guardrails against AI use in future elections.
The 2026 Brazilian national election proves to be a fascinating case study of this issue. This year, the nation’s Superior Electoral Court (TSE) had to decide how the electoral law would apply to AI-generated political content.1 Its decision has begun a new conversation surrounding how democratic institutions must safeguard from abuse of new technologies. Simultaneously, it has raised questions surrounding the implications of giving these institutions more control over political communication.
The Brazilian case ultimately suggests that democracies must be protected from AI manipulation of content and communication, but regulations must be carefully constructed so as to create clear and consistent standards to avoid placing any political candidates or parties at a disadvantage.
How Has AI Infiltrated Brazilian Elections?
On September 1 of this year, Brazil’s TSE rejected a request that sought to fine Flávio Bolsonaro, a presidential candidate, over an AI-generated video that was shown of his father, former President Jair Bolsonaro, endorsing his son during a party convention. The court decided that broadcasting this content at a party convention, by itself, did not classify as propaganda. Their decision further stated that the content and context of such messages were crucial when determining if they should be classified as propaganda. Furthermore, the court simultaneously released criteria for identifying deepfakes and clarified when prohibition of content shall be permitted.2
Yet, the court took additional actions in October by removing three campaign posts that had been published by Flavio Bolsonaro. One of these posts had utilized AI to artificially depict President Luiz Inácio Lula da Silva in a fabricated video. The court declared that since the post did not sufficiently disclose that the content was AI-generated, it could not be utilized by political actors.3
These respective decisions are contrasting in nature, showing the TSE’s willingness to allow artificial intelligence to be incorporated into political content in certain contexts while simultaneously drawing the line on its use in other respects. From this, the question emerges of how an institution such as the TSE implements such regulations without abusing its institutional discretion.
Where Does Democratic Erosion Fit In?
Drawing on Waldner and Lust’s work, democratic backsliding refers to the deterioration of institutions and practices which would otherwise sustain democracy.4 This idea is relevant in the Brazilian case, as the emergence of deepfake content in campaigns does not necessarily contribute to democratic erosion directly, yet the intervention by the TSE does not appear to be making institutions more democratic either.
Other scholarship has offered additional frameworks that offer an approach to analyzing this dilemma. Baron et al. promote the use of an events-based framework to indicate the quality of a democracy at any given time as opposed to relying on more broad, long-term changes.5 The TSE’s decisions are not concrete indicators of change in how democratic Brazil is, instead, they serve as smaller-scale events that reveal how institutions respond to new threats. These may be fairly insignificant in the short term, but can prove incredibly useful when determining how the conditions for political competition have been preserved over longer periods of time.
Who is Really Accountable in Brazil?
Another lens to examine these recent threats to Brazilian political content is that of horizontal accountability. O’Donnell’s “Delegative Democracy” makes the distinction between accountability via elections and accountability through institutions with large amounts of autonomy. His argument asserts that in a representative democracy, institutions like courts, legislatures, and others must have the bandwidth to challenge or question the actions of those in power.6
Intervention by the TSE in Brazil is an example of this logic in practice. Political candidates have been denied the right to determine whether their actions during campaigns are compliant with the rules and norms that precede them. As an independent electoral institution, the TSE is allowed to intervene when they feel candidates have crossed a line, with the idea that voters will be more protected in the process.
Yet, O’Donnell’s theory also demands that institutions be evaluated on how they operate, as their existence alone does not necessarily ensure democracy will be upheld.6 This provides an interesting consideration in the Brazilian case. The TSE is tasked with monitoring political actions to protect voters from deepfakes or other AI content that may be misleading while ensuring they do not restrict speech to a degree that could constitute an abuse of its institutional power. Leveraging these two democratic interests is difficult in this case, and is why the court released two fairly contrasting decisions in September and October.
When Safeguards Become Anti-Democratic
Formal enforcement mechanisms as seen with the TSE’s exercise of power in this case have long been scrutinized by scholars. Varol’s theory of stealth authoritarianism suggests that institutions can exercise their power to a degree that weakens democratic competition in a given society.7 To some, the power used by the TSE to regulate campaign content could be deemed an undemocratic consequence of a democratic institution.
The counterargument to this is strong. If the TSE had not intervened, candidates would be free to use AI-generated content, which could fabricate information about their opponents or give voters a false impression of reality. Requiring the disclosure of AI use and removal of fabricated videos may be a necessary move to ensure elections are secure. Still, this argument does not eliminate the need for regulations of the institutions that impose such restrictions. Enforcement of rules is important, but it is not necessarily democratic if it lacks consistency in reasoning and isn’t applicable across competing political parties. Electoral oversight has been allowed in this instance, but it must also be allowed in different elections with perhaps less notorious political candidates.
How Can Brazil Ensure Democratic Resilience?
Brazil’s experience is a perfect illustration of how democratic erosion cannot be understood by simply examining the most impactful political controversies or major institutional overhauls. The TSE represents the necessity of such regulatory institutions when a society faces a technological threat to its elections. However, it cannot be recognized as a democratic institution if it does not exert authority consistently across political conflicts. The challenge here is not to make a choice between the regulation of political content and the protection of free speech in elections. It is rather to ensure there are established protocols that can allow both without favoring a political ideology or undermining any emerging competition.
AI has made this balance a much harder line to toe. A resilient democracy is contingent on strong institutions that enforce clear electoral rules. Still, such institutions must be held accountable enough to protect democracy and assure society that they are not abusing their power.
Endnotes
Ricardo Brito and Luciana Novaes Magalhaes, “Brazil’s Bolsonaro Is Barred from This Year’s Election. Can His AI Avatar Campaign Instead?,” Reuters, July 28, 2026, updated July 29, 2026, https://www.reuters.com/business/media-telecom/brazils-bolsonaro-is-barred-this-years-election-can-his-ai-avatar-campaign-2026-07-28/.
Tribunal Superior Eleitoral, “TSE fixa tese sobre deepfake e delimita regra para as Eleições 2026” [TSE establishes criteria for deepfakes and defines rules for the 2026 elections], September 1, 2026, https://www.tse.jus.br/comunicacao/noticias/2026/Setembro/tse-fixa-tese-sobre-deepfake-e-delimita-regra-para-as-eleicoes-2026.
Tribunal Superior Eleitoral, “TSE manda retirar do ar vídeos com uso de inteligência artificial publicados por Flávio Bolsonaro” [TSE orders removal of videos using artificial intelligence published by Flávio Bolsonaro], October 8, 2026, https://www.tse.jus.br/comunicacao/noticias/2026/Outubro/tse-manda-retirar-do-ar-videos-com-uso-de-inteligencia-artificial-publicados-por-flavio-bolsonaro.
David Waldner and Ellen Lust, “Unwelcome Change: Coming to Terms with Democratic Backsliding,” Annual Review of Political Science 21 (2018): 93–113, https://doi.org/10.1146/annurev-polisci-050517-114628.
Hannah Baron, Robert A. Blair, Jessica Gottlieb, and Laura Paler, “An Events-Based Approach to Understanding Democratic Erosion,” PS: Political Science & Politics 57, no. 2 (April 2024): 208–15, https://doi.org/10.1017/S1049096523001026.
Guillermo O’Donnell, “Delegative Democracy,” Journal of Democracy 5, no. 1 (January 1994): 55–69, https://www.journalofdemocracy.org/articles/delegative-democracy-/.
Ozan O. Varol, “Stealth Authoritarianism,” Iowa Law Review 100, no. 4 (2015): 1673–1742, https://ilr.law.uiowa.edu/print/volume-100-issue-4/stealth-authoritarianism.

0 Comments