Brazil Tried to Regulate AI in Elections. A Bolsonaro Avatar Is Testing the Rules

Deepfakes pose a significant challenge for Brazil as they prepare for upcoming elections, as a newly created Jair Bolsonaro avatar is being utilized in campaign content, revealing gaps in existing regulations. The Superior Electoral Court (TSE) is grappling with questions surrounding the usage of AI-generated content in political campaigns, specifically in the case of an AI-generated version of former President Jair Bolsonaro being used to support his son Flávio Bolsonaro's presidential campaign.

The Bolsonaro avatar's presence raises a legal question - can an AI-generated likeness effectively convey a message when the real person cannot due to restrictions? This case arrives as AI-generated political content grows online, and more voters utilize chatbots seeking information about candidates. Despite Brazil having explicit election AI rules, enforcing them appears increasingly complex.

The issue is not the use of AI in a campaign, but the subtlety with which AI can cross established boundaries without appearing to do so. The TSE cleared Flávio Bolsonaro's campaign on electoral grounds, yet Supreme Court justices continue to evaluate whether deploying an avatar by a third party violates the conditions of Jair Bolsonaro's sentence.

The legal interpretation lies in whether the synthetic version allowed his political message to reach voters through a machine-generated likeness without his direct authorization.

Under the TSE's rules, deepfakes are banned, including synthetic content used to alter or create a person's image or voice for the purpose of spreading false information or harming a candidacy. AI systems are also prohibited from ranking or recommending candidates, parties, or campaigns, even when asked by a voter. The court has also tightened the rules around election day, prohibiting AI-generated or altered content during the 72 hours prior to voting and the 24 hours following the election.

Brazil is requesting providers adhere to its AI laws and submit compliance plans during the election period.

This case highlights the challenge of regulating AI in elections, where the use of AI itself is permissible but not unlimited. As AI chatbots become more popular, attackers can manipulate the information these systems retrieve and summarize. A February demonstration showed how AI tools could surface a fake blog post, suggesting that political actors could use manipulated web content to influence voters. Brazil's rules prohibit AI systems from ranking or recommending candidates, but popular tools are already doing so.

Generative AI has the potential to increase the volume of political misinformation produced by a single campaign, making it harder for platforms and fact-checkers to identify coordinated misinformation. As governments grapple with these issues, they may require AI providers to explain how political queries are handled, restrict candidate recommendations, create faster processes for challenging synthetic campaign content, or impose tighter rules in the weeks leading up to voting.

For voters, understanding the complexity of AI usage in elections may go beyond identifying deepfakes, as they may need to question the sources of AI-generated political information and whether the person on screen ever existed in that form.