Brazil’s Electoral Court Tightens Internet and AI Rules for 2026 Vote

Brazil · POLITICS

Approved in March and tested within days of the campaign opening, the electoral court’s new rulebook redraws Brazil’s online election weeks before the October 4 first round.

BRASÍLIA, Wednesday, August 26, 2026 — With Brazil’s official campaign already on the streets and free radio and television time starting on Friday, the tougher internet regime adopted by the Superior Electoral Court (TSE) for the 2026 elections has begun its first stretch of real enforcement. Resolution 23.755, approved by the court’s plenary on March 2 and now the binding rulebook for the October vote, keeps an absolute ban on electoral deepfakes, requires explicit labeling of any campaign content produced or manipulated by artificial intelligence, creates a blackout on new synthetic material around voting days and hands digital platforms fresh compliance duties. The Brazil election AI rules met their first high-profile case within days of the campaign opening: on Tuesday, Justice André Mendonça ordered the removal of an Instagram video aimed at presidential candidate Flávio Bolsonaro and proposed a legal test for what counts as a deepfake.

What Resolution 23.755 changed

Resolution 23.755 of March 2, 2026 amends Resolution 23.610 of 2019, the standing rulebook on electoral propaganda. The text was drafted by Justice Kassio Nunes Marques — then the court’s vice-president and now its president — after an internal working group circulated drafts and received roughly 1,500 suggestions from parties, civil society and regional electoral courts, according to Valor Econômico’s coverage of the vote. The plenary concluded its approval on Monday, March 2, and the resolution was published within the legal deadline that week. The Brazil election AI rules apply to parties, candidates, federations and coalitions — and, more heavily than ever, to the platforms themselves.

The core of the package is Article 9º-B. Anyone responsible for electoral propaganda who uses synthetic multimedia content generated by AI — to create, replace, omit, merge or alter images or sounds — must inform voters in an explicit, prominent and accessible manner that the material was fabricated or manipulated, and must name the technology used. In printed material, the notice has to appear on every page or face that uses AI content. A paragraph added this year obliges boosting providers to offer a specific field in which advertisers declare the use of artificial intelligence, closing a loophole that let paid posts circulate without any trace of how they were made.

Deepfake ban and the 72-hour blackout

The resolution keeps the outright ban, introduced in 2024, on deepfakes — audio or video generated or digitally manipulated to create, replace or alter the image or voice of a living, dead or fictitious person — when used to harm or favor a candidacy. The prohibition holds even with the consent of the person depicted and even if the content carries a label. In May, the full bench unanimously upheld a R$15,000 fine against Fortaleza mayor Evandro Leitão over a 2024 TikTok deepfake, holding that digital manipulation with electoral purpose is sufficient to establish the offense, regardless of whether voters could actually be deceived. Penalties range from content removal and fines of R$5,000 to R$30,000 (about US$970 to US$5,800) to cancellation of a candidacy registration or loss of office.

The novelty with the greatest practical impact is paragraph 3º-A of Article 9º-B: a blackout window. In the 72 hours before each round of voting and the 24 hours after polls close, no new synthetic content using the image, voice or statements of candidates or public figures may be published, republished or boosted — even if correctly labeled. Lawyers advising campaigns describe the window as the court’s answer to the classic last-minute smear, when a viral fake leaves no time for fact-checks or a right of reply. The Brazil election AI rules also treat sexually manipulated imagery of candidates as potential gender-based political violence under Law 14.192 of 2021, stacking electoral, civil and criminal liability on the same conduct.

Platform duties and advertising transparency

Under the Brazil election AI rules, platforms carry a heavier load than in any previous cycle. Article 125-B requires application providers to file compliance plans with measurable duties and deadlines to protect the integrity of the vote. A new paragraph in Article 28 forbids providers offering AI systems — even at the user’s explicit request — from ranking, recommending or prioritizing candidates or parties, from issuing electoral opinions or vote recommendations, and from creating sexual, nude or pornographic alterations of images of candidates.

On paid advertising, boosted content must be clearly identified as such, only candidates, parties, federations and coalitions may contract it, and platforms offering boosting must register with the electoral justice and keep public repositories of paid political content, including whether AI was used. Judges have been ordering removals within 24 hours of notification, and the resolution makes providers jointly liable, in the civil and administrative spheres, if they fail to immediately take down notified content in the risk categories it lists, including unlabeled synthetic material and deepfakes. The resolution also extends profile-removal powers to accounts engaged in repeated electoral crimes or the dissemination of notoriously false facts, and the mass messaging blasts banned in 2024 remain prohibited.

How the court voted — and who pushed back

The 2026 resolutions were approved by the full bench in sessions that ended on March 2, after public hearings in February. Closing those hearings, Nunes Marques promised that digital campaigning would face rigorous monitoring, with the court acting neither with excess nor with inaction. Not everyone is satisfied, and the pressure runs in both directions. The Justice Ministry’s representative at the hearings, Ricardo de Lins Horta, argued for even broader removal powers, warning that limiting profile bans to proven criminal accounts could grant immunity to profiles with immense potential to unbalance the vote.

From the other side, lawyers and scholars writing in Consultor Jurídico (ConJur) have attacked two innovations: Article 9º-I, which lets judges reverse the burden of proof in AI disputes when verifying manipulation is excessively costly for the complainant, and the blanket ban on AI systems recommending candidates, which one commentator described as the court trading the scalpel of 2024 for a guillotine. Enforcement data feed the skepticism: news portal Terra reported this month that of 591 deepfake-related decisions in the 2024 municipal elections, only about 20 percent recognized manipulation, while 62 percent treated the content as satire or humor.

Brazil election AI rules face their first real test

The rules are no longer abstract. Mendonça’s Tuesday injunction — issued in a case over a video targeting Flávio Bolsonaro — is the first high-profile application of the Brazil election AI rules in the 2026 cycle, and his proposed realism threshold, which would reserve the deepfake label for content realistic enough to create a false perception of authenticity, still has to go before the plenary. The Rio Times reported that proposal this morning, after earlier coverage of the court weighing a deepfake of Jair Bolsonaro shown at the PL convention and of the first disputes over how the new rules apply.

More tooling is on the way. Legal-news outlet JOTA reported that court technicians are studying AI-detection software for use this cycle, on the understanding that probabilistic detectors should inform — but never solely decide — judicial rulings. Since July, the court’s public repository of disinformation rulings, known as Reade, has grouped removal decisions into thematic axes, including one dedicated to irregular AI use. The calendar is tight: campaigning opened on August 16, free airtime on radio and television runs from August 28, and Brazilians vote on October 4, with an October 25 runoff where needed. However the coming cases land, 2026 is the first general election in which the Brazil election AI rules apply at full national scale.

Frequently Asked Questions

Sources

• valor.globo.com
• www12.senado.leg.br
• www.conjur.com.br
• www.conjur.com.br
• www.terra.com.br
• www.tre-sp.jus.br
• www.jota.info

This article was produced by The Rio Times’ automated newsroom system. How we use AI · Report an error