Brazil's Election Court Races to Contain AI Before October Vote

Imagine scrolling through your social feed and seeing a former president, one who is legally barred from speaking to the public, addressing a crowd anyway. Not through a loophole in the law, but through a digital stand-in built to look and sound just like him. That is not a hypothetical for Brazilian voters. It happened in July, when an AI avatar of former President Jair Bolsonaro, who is under house arrest and prohibited from public communication, appeared at his son's party convention. The episode triggered immediate legal challenges and became a preview of the messier fight now playing out ahead of Brazil's October 4 election.

Brazil's top elections court, known as the TSE, is expected to rule Tuesday on that avatar case, and with it, set clearer limits on AI-generated or altered content in campaigns. The stakes are not abstract. Voters go to the polls in just over a month, and the digital campaign has already been reshaped by tools that can generate convincing video, audio, and images faster than regulators can review them.

Think of the TSE's task like trying to patch a boat while it is already crossing the river. The court rolled out expanded AI restrictions for this cycle: mandatory labels on AI-generated content, bans on deepfakes, prohibitions on AI systems recommending candidates, and a blackout period barring new AI-generated material right before and after voting. Tuesday's ruling is expected to sharpen one of the murkiest pieces of that framework, offering a clearer legal definition of what actually counts as a deepfake.

That clarity matters because so much of what is circulating sits in a gray zone. Instituto Democracia em Xeque tracked hundreds of social media posts from politicians aligned with Bolsonaro and President Luiz Inácio Lula da Silva that were formally labeled as AI-generated over just 15 days. Some of it is genuinely playful: a right-wing gubernatorial candidate in Amazonas state used AI to drop her ballot number into different scenic backdrops, and a Lula ally shared a video appearing to dance alongside the president. Bolsonaro's campaign has leaned into the format harder than Lula's, sharing an AI-produced campaign song, a video of himself and his running mate as dancing dolls, and an AI animation depicting him as a chiseled figure in a swim brief that quickly spawned a wave of memes.

Where the Real Danger Lies

None of that, on its own, threatens democratic integrity. The bigger worry, according to João Victor Archegas, legal coordinator at the Institute of Technology and Society (ITS) of Rio de Janeiro, is content designed to convincingly depict events that never happened.

"It is not the tool itself that is the problem," Archegas said. "It is its potential use in certain contexts that may lead to dangerous outcomes for our democracy."

That distinction, harmless novelty versus fabricated reality, is exactly what the TSE is trying to draw a firmer line around. The court has already fielded 19 cases involving AI use in just the first 11 days of this year's campaign, according to court data. One case in early August involved a video showing Bolsonaro alongside Daniel Vorcaro, the former banker at the center of a major fraud scandal, in scenes that never actually occurred. TSE Justice André Mendonça ordered it removed.

The scale of AI's reach into Brazilian political life goes well beyond flashy videos. A survey by the Fundação Itaú Observatory found that AI is already part of daily life for 75% of Brazilians, which means the technology is not some fringe curiosity voters encounter occasionally. It is showing up in fake testimonials about candidates and fabricated polls, quietly shaping impressions before anyone fact-checks a thing.

Chatbots present a subtler version of the same problem. A May survey by Projeto Brief found that nearly 63% of voters were willing to consult AI tools about candidates, treating them almost like a trusted, neutral source. The TSE's rules explicitly ban generative AI systems, including tools like ChatGPT, from recommending or ranking candidates, even when a user directly asks for that. But a study by ITS Rio found that seven of the most widely used chatbots in Brazil ranked or ordered candidates in 90% of their responses anyway. Writing a rule is one thing. Getting the technology to actually follow it is another matter entirely, and that gap is where much of the risk now lives.

It is worth pausing on why that gap exists. Chatbots are built to be helpful and responsive, trained to give users something resembling an answer even when the honest answer is "I shouldn't weigh in on this." That design tension, between being useful and being neutral, does not disappear just because a court issues a rule against it. Regulators can write the law, but enforcing it against a system that generates new responses every time someone asks a question is a fundamentally harder task than banning a single video or image.

What Comes Next

Tuesday's ruling will not settle the matter outright. Watchdog groups have already suggested the TSE's existing regulations are having a limited impact on the flood of content reaching voters, and a tighter legal definition of a deepfake, however useful, will not by itself stop chatbots from ranking candidates or prevent the next fabricated video from spreading before anyone can flag it.

What the ruling can do is give the court sharper tools for the cases still to come, and there will likely be more of them before October 4. Brazil's experience offers a preview for other democracies watching how AI is starting to reshape political discourse, exposing how technical rules only carry as much weight as the enforcement behind them. The lesson so far is a sobering one: writing good policy is the easier half of the job. Making the technology actually comply with it, especially in the final stretch before an election, is proving to be the harder fight.