When more than 155 million Brazilians head to the polls on October 4, they will do so under what legal scholars call one of the world’s most detailed regulatory frameworks for artificial intelligence in elections. Brazil election AI rules approved unanimously by the Superior Electoral Court (TSE) in early March ban deepfakes outright, require mandatory labeling of all AI-generated campaign material, prohibit AI tools from recommending candidates to voters, and impose a total blackout on synthetic content near election day. Yet the same experts who praise the ambition warn that critical gaps — around AI developers, private messaging, and enforcement penalties — could leave the door open to the very manipulation the rules are designed to prevent. This is part of The Rio Times’ daily coverage of Brazil financial news English and Latin American financial markets.

Brazil Election AI Rules: What Changes

The framework builds on rules first introduced for the 2024 municipal elections and expands them significantly for the first general election under these provisions. The core prohibition bars publication, republication, or boosting of any AI-altered content on social media from 72 hours before voting until 24 hours after polls close. Any campaign material using AI to create, modify, merge, or overlay images or audio must carry a clear, prominent, and accessible disclosure. Deepfakes — synthetic content manipulating a real person’s voice or likeness — are banned entirely in electoral propaganda. A candidate who uses deepfakes risks having their registration or mandate annulled. The TSE also explicitly prohibited AI systems from ranking, recommending, or prioritizing candidates, even if a user requests a suggestion.

Platforms face joint civil and administrative liability if they fail to remove illegal content “immediately” after notification during the campaign period. In a significant procedural innovation, the rules allow the burden of proof to be reversed in AI-related election lawsuits: if a party or candidate challenges the authenticity of content and manipulation is suspected, the publisher may be required to prove the material is genuine, rather than the accuser having to prove it is fake. The resolution also introduced specific protections against AI-generated sexual or pornographic deepfakes targeting candidates, reinforcing gender-based political violence legislation already on the books.

Where the Gaps Are

Fernando Neisser, an electoral law professor at FGV São Paulo, identified the most significant blind spot: the rules concentrate enforcement on social media platforms but say almost nothing about the companies that build the AI systems generating manipulated content in the first place. “If the user can’t identify it, the courts can’t identify it, and even the Federal Police struggle — how can we expect Instagram to detect it?” he asked. Without obligations on AI developers to implement watermarking or detection tools, the burden falls on platforms that may lack the technical capacity to comply.

Private messaging environments represent another unresolved challenge. WhatsApp and Telegram are the primary channels through which political content circulates in Brazil, yet the resolution does not clearly define when AI-generated material shared in these spaces crosses from private conversation into regulated electoral propaganda. Researcher Erick Beyruth of PUC São Paulo warned that the current text “creates the impression that these environments are also immune to electoral rules,” opening a significant gray zone. Electoral attorney Alberto Rollo added that while the reversed burden of proof is a positive step, identifying AI manipulation with certainty remains technically complex even with forensic analysis, meaning some disputes may be unresolvable within campaign timelines. As The Rio Times has reported, leading campaign strategists expect 2026 to be “the year of AI” in Brazilian politics — and the question now is whether the rules can keep pace with the technology they aim to regulate.