Brazil Electoral Court Decides If Bolsonaro AI Video Is a Banned Deepfake

BRAZIL · ELECTIONS 2026

Five weeks before the first round, Brazil’s electoral court is about to draw the line between legal AI content and an illegal deepfake — in a case starring Jair Bolsonaro, a father barred from campaigning who appeared on a convention screen anyway.

The video that forced the question

On 25 July, at the national convention that confirmed Senator Flávio Bolsonaro as the Liberal Party’s presidential candidate, the party played a video in which former president Jair Bolsonaro — under house arrest and barred from political statements by Supreme Court Justice Alexandre de Moraes — appears to declare support for his son. The figure on screen was generated by artificial intelligence, and the event was streamed live on YouTube.

The Workers’ Party (PT) took the case to the Superior Electoral Court the next day, arguing that the recording uses technology to distort image and voice and therefore fits the definition of a deepfake, which the court’s resolutions ban from electoral propaganda.

Flávio Bolsonaro’s campaign answered that the video broke no law because it never hid the use of the technology — the piece was openly labeled as AI-made. The defense of the former president added that he did not know the video would be produced or shown at his son’s launch event.

The case lands on the desk of the court’s president, Justice Kassio Nunes Marques, who is its rapporteur and has told colleagues he wants the ruling to set a precedent for future episodes. Judgment is scheduled for Tuesday evening, 1 September.

A court split over what a deepfake is

Inside the TSE, two readings compete. One wing sees little gravity in the episode: since the video carried an AI warning, there was no attempt to deceive the voter. Another current, led by Justice Estela Aranha, has argued for the removal of all deepfake content from the electoral context, labeled or not.

Last week, the court’s vice-president, Justice André Mendonça, proposed objective criteria to close the gap: content should be treated as a deepfake when its realism is capable of producing a false perception of authenticity in the viewer — a test that looks at the effect on the audience, not at the intent of the producer.

The political context adds voltage. Because Jair Bolsonaro is prohibited from manifesting himself during the election period, Supreme Court Justice Flávio Dino published a text after the convention drawing a parallel between totalitarianism and trickery — widely read in Brasília as a warning that the AI video may have been an end-run around Moraes’s ban.

The rules already in force

Whatever the verdict, Brazil’s 2026 campaign already runs under the strictest AI rulebook in the country’s electoral history, approved by the TSE in early March and effective for the entire general election.

Every campaign piece that uses AI to create or alter image, voice or content — printed or digital — must carry an explicit, prominent and accessible warning, and must say which technology was used. Avatars and chatbots talking to voters must identify themselves as machines.

In the 72 hours before each round and the 24 hours after it, publishing or boosting any new AI-produced or AI-altered content about candidates is absolutely banned — the court’s answer to the classic “election-eve slander” problem, now supercharged by generative tools.

The rules also reach the technology industry. AI providers may not recommend candidacies, even when users ask them to — a clause designed to prevent algorithmic interference in the voter’s decision. Platforms that fail to remove unlabeled or prohibited synthetic content immediately share liability and face fines and other sanctions. Sexualized deepfakes against female candidates are treated as gender-based political violence under a 2021 law.

Sanctions scale from immediate takedown to fines and, in the gravest cases, cancellation of the candidacy registration or loss of an elected mandate.

The caseload is already here

The pre-campaign has kept the three justices assigned to electoral propaganda — Nunes Marques, Mendonça and Aranha — busy. Both the PT-led federation and the PL have won emergency orders forcing platforms to take down posts within 24 hours.

In one case, Aranha ordered Instagram to remove a Flávio Bolsonaro video associating PT lawmaker Gleisi Hoffmann with “violence and criminality,” treating it as early negative campaigning. In another, the TSE upheld a R$5,000 (about US$965, at roughly 5.19 reais to the dollar on 1 September) fine issued by the Paraná regional court over an AI video spread in a WhatsApp group during the 2024 municipal elections.

Behind the scenes, court technicians are studying AI-detection tools for use this cycle, in house or from the market. The working understanding, reported by the legal outlet JOTA, is that detectors work on probability and never reach full accuracy — so their output can inform a ruling but never decide one alone.

What comes next

If the court classifies the convention video as a deepfake, the PL risks sanctions and every campaign inherits a narrow definition of what AI can do in the final five weeks. If it accepts the labeling defense, labeled synthetic endorsements become a lawful — and certainly copied — campaign tool.

Either way, the first round on 4 October will be the first national stress test of the framework built in 2024 and hardened in March. The ruling expected on Tuesday night tells campaigns exactly how much room they have left.

Frequently Asked Questions

Sources: Poder360; g1; CNN Brasil; Valor Econômico; Estadão; JOTA; Gazeta do Povo; TSE resolutions 23.610/2019, 23.732/2024 and the March 2026 amendment. FX basis: USD/BRL 5.1866, Investing.com, 1 September 2026.

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