Brazil's electoral court judge orders pro-Lula AI video pulled

A judge on Brazil's Superior Electoral Court ordered Jorge Messias, the head of the federal government's legal office, on 13 October 2026 to remove a pro-Lula video made with artificial intelligence from his Instagram account within 24 hours, after the campaign of Lula's runoff rival complained the clip carried no AI warning. The order landed 12 days before the 25 October runoff.

The preliminary order came from Marco Aurélio Bellizze, a substitute minister of the seven-member court and, since 7 October, its auxiliary judge for campaign advertising in the 2026 general election, according to the TSE. The order also binds Meta, the California company that owns Instagram, and bars the video from returning without the warning the rules require, Correio Braziliense and Agenda do Poder reported.

Messias is the Advogado-Geral da União, the head of the AGU, the federal government's legal arm, with the rank of a cabinet minister. Bellizze wrote that the labeling rule does not turn on how convincing the fake looks: "The rule does not make removal depend on how realistic the content is, nor on proof that the voter was actually misled," he wrote, in a translation of the Portuguese text published by Correio Braziliense and Agenda do Poder. In the same ruling, as carried by SBT News, he described the clip's figures this way: "The old man, the child, the setting and the voices do not correspond to the record of a real scene."

The video went up on Messias's Instagram account on Saturday 10 October. According to Correio Braziliense and Agenda do Poder, it shows an AI-made elderly man and a child in red shirts printed with "Lula 13", the president's ballot number, preparing for a car rally and repeating the number. Senator Flávio Bolsonaro of Rio de Janeiro, the president's opponent in the runoff, took the case to the court on Monday 12 October; his lawyers argued the scene never took place and the people never existed, and that the clip never told viewers it was synthetic, according to The Rio Times. They asked for a fine of R$5,000 to R$30,000 (about US$1,000 to US$6,000, at 5.02 reais to the dollar), The Rio Times reported.

Bellizze granted the request in part. He left the fine, and the question of whether the clip is a banned deepfake, to the court's final ruling, and sent the case to the full court.

The same rulebook cut against Lula's rival five days earlier. On 8 October 2026 the TSE ordered, unanimously, that Flávio Bolsonaro remove three posts from X within 24 hours, under penalty of a daily fine, according to the court. One used AI to simulate Lula's image in a situation that did not happen and was not labeled as AI; the other two, according to the court's account of the complaint, associated Lula with the criminal groups PCC and Comando Vermelho. The decision answered a request by Federação Brasil da Esperança, an alliance led by Lula's Workers' Party; the reporting minister, André Mendonça, reversed an earlier position to grant it, and the platform was to be notified.

The rule behind both orders is Resolution 23.610/2019, the TSE's campaign rulebook. Article 9-B, added in 2024, requires anyone who uses AI-generated or AI-manipulated images or sounds in campaign material to say so explicitly, prominently and accessibly and to name the technology; for video, the warning must run at the start and as an on-screen label. Article 9-C bans deepfakes, defined as synthetic audio or video that creates, replaces or alters the image or voice of a living, deceased or fictional person to help or hurt a candidacy. Breaking that ban, the resolution states, counts as abuse of political power and misuse of the media, and can cost a candidate a registration or a mandate.

The United States handles the same problem state by state. The National Conference of State Legislatures counted 31 states with election deepfake laws in a June 2026 report, 28 of them requiring disclosure, according to a Governing analysis published on 9 October 2026. In many of those states, a warning is enough and the ad can stay up; Maryland, Minnesota and Texas prohibit certain election deepfakes even when the content carries an AI warning, Governing reported. A federal judge temporarily barred Montana from enforcing its deepfake law against a conservative PAC in September 2026, Governing reported.

For the people who make and post AI video, the Brazilian case shows how quickly the label becomes the line. A single substitute judge can order a campaign post off a US-owned platform within a day, and the reasoning covers fictional characters as much as real ones, so a synthetic scene built to promote a candidate draws the warning duty whether or not it depicts an actual person.

The seven-member court now reviews Bellizze's preliminary order, a step the TSE's own rules require. Its final ruling will set any fine and decide whether the clip is a prohibited deepfake.

Brazil votes in the runoff between Lula and Flávio Bolsonaro on Sunday 25 October 2026.