Brazil's electoral court has ordered the Liberal Party to take down an AI-made Instagram video that links President Lula to a criminal faction

Key takeaways

• The vote was unanimous on removal. The grounds were not: the rapporteur relied on admissible-criticism limits, while two ministers relied on the use of AI to simulate the pre-candidate.
• The video is described in the release as produced with artificial intelligence and as simulating a handshake between Lula and a former federal deputy inside a prison.
• The source is a TSE news release, not the judgment. The rapporteur's quoted reasoning is expressly an analysis made "em análise preliminar".

What did the TSE decide?

According to the TSE news release, the ministers of the Tribunal Superior Eleitoral determined that the Partido Liberal remove a video posted on Instagram that associates President Luiz Inácio Lula da Silva, a candidate for re-election, with a criminal faction. The release says the decision was unanimous and was taken by the Plenary in a judgment session on the night of Tuesday 13 October.

The release describes the video as produced with artificial intelligence. It says the video links Lula to former federal deputy Uldurico Junior (MDB-BA), who was arrested in a Federal Police operation on suspicion of having received R$ 2 million to facilitate the escape of 16 detainees from a prison in Bahia. It adds that Uldurico ran for mayor of Teixeira de Freitas (BA) in the 2024 municipal elections, when Lula supported him. The release lists a related case, Representação 0600695-85.2026.6.00.0000. As to the content, it says the video simulates a handshake between Lula and Uldurico inside a prison unit, reproduces an excerpt from a news report mentioning "uma aliança com lideranças das facções criminosas", and shows a cell with a faction's initials on the wall and Lula's image in the background.

Why did the rapporteur find the video unlawful?

The release reports that Minister Estela Aranha, the rapporteur, recalled that TSE Resolution 23.610/2019 provides for minimal intervention by the Electoral Courts in the digital environment, and that removal of content is nonetheless admitted where electoral rules are violated or participants in the electoral process are offended.

Her reported vote reads: "A publicação impugnada, em análise preliminar, extrapola os limites da crítica política admissível", in that it attributed to the pre-candidate generic links with criminal organisations without minimal factual support, which she called gravely disinformative content. In English, the vote says the challenged publication, on a preliminary analysis, goes beyond the limits of admissible political criticism. She added, according to the release, that TSE case law treats the association of candidates or parties with criminal organisations without a factual basis as a violation of honour that can amount to irregular electoral advertising. The release does not quote the operative paragraph of the judgment.

How did Ministers Floriano de Azevedo Marques and André Mendonça differ?

Minister Floriano de Azevedo Marques followed the rapporteur on removal but diverged on the grounds. The release says that, for him, the material used AI to simulate the human figure of the pre-candidate, a practice barred by the electoral rules, and that reference to a candidate's support for a politician later accused of organised-crime involvement is not, by itself, an irregularity that justifies removing the advertising.

He is quoted: "Porém, como nessa publicação há o uso da inteligência artificial vedado no artigo 9º-C, parágrafo 1º, da Resolução-TSE 23.610/2019", adding that he joined the rapporteur's operative conclusion with different reasoning. The quoted passage places the unlawful AI use in article 9-C, paragraph 1, of Resolution 23.610/2019. Minister André Mendonça, following the divergence, is quoted saying "A informação aqui em si não é falsa." and "O que traz a ilicitude para essa propaganda específica e faz cair toda a propaganda é o indevido uso da inteligência artificial" (the full sentence continues with the words "à luz da jurisprudência que consolidamos"). In English, he said the information is not false and that the improper use of AI is what makes this advertisement unlawful.

The release does not state how many ministers joined each ground, and does not report the votes of the remaining ministers beyond the unanimous outcome.

What should election lawyers, campaign teams and platforms do now?

The release shows two separate removal routes in one case. The steps below are our editorial analysis, not requirements stated in the release or imposed by the Court.

• Editorial analysis: check content claims first: any AI-assisted post that ties a candidate or party to a criminal organisation needs a concrete factual basis, since the rapporteur's reported reasoning turned on its absence.
• Editorial analysis: check the AI use independently: if a post simulates a real candidate's human figure with AI, assess it against article 9-C of Resolution 23.610/2019 even where every factual statement in it is accurate.
• Editorial analysis: keep the generation record: save the prompts, tools and approvals behind any AI-made political video, so the party can show what was produced and why.
• Editorial analysis, for platform counsel: obtain the judgment text before building a takedown rule on either ground, because the release does not say which of the two grounds the Court as a whole adopted.

What are the key dates and parties?

What we did not verify

What we opened: the TSE news page on tse.jus.br on 25 October 2026, including the quoted passages from the votes of Ministers Estela Aranha, Floriano de Azevedo Marques and André Mendonça.

What we did not open: the judgment or the case file, the text of Resolution 23.610/2019 and its article 9-C, the video itself, the Federal Police operation, and any statement from the Partido Liberal. Every statement here is limited to what the TSE release said; we did not independently examine the judgment, the resolution, the video or the police evidence. We do not know the removal deadline, whether review has been sought, or which ground the full Court adopted.

What we refuse to claim: that the Court ruled AI-simulated candidate depictions unlawful as a general rule, that the video's content was false, or that the Federal Police allegations against Uldurico Junior are established. The release reports them as suspicion.

Related briefings

• AI Regulation News hub
• Latin America AI regulation tracker
• Attorneys and AI regulation
• TSE election deepfake ban and AI labelling rules
• TSE and AGU accord on AI good practices in elections
• STF Justice Moraes order on an AI avatar and electoral deepfake
• Brazil PL 2338 AI framework bill status