TSE Orders an Unlabelled AI Video Removed and Gives Meta and X 24 Hours to Explain Their Systems Before the Runoff
Brazil's Superior Electoral Court (TSE) acted twice this week on artificial intelligence in the presidential race, days before the runoff on Sunday, October 25. Minister Marco Aurélio Bellizze ordered the removal of an AI-assisted video from “Tropa do Flávio,” a WhatsApp channel run by Senator Flávio Bolsonaro's campaign with about 2.1 million followers, CNN Brasil reported on October 23. Separately, the court said its electoral inspector general had given Meta and X 24 hours to explain how their systems treat the candidates.
The video, 1 minute and 17 seconds long, posted on October 14 and since deleted by the channel, opened with a figure resembling President Luiz Inácio Lula da Silva seen from behind, facing a Brazilian flag covered in red smoke, chains and debris, followed by unsourced statistics on debt, hunger and business closures. Lula's coalition argued the opening image was generated by AI without the label the law requires. In a preliminary injunction that partly granted the request, Bellizze found the claim of unlabelled synthetic content plausible, called the piece offensive and misleading, and cited the risk of harm in the final stretch of the campaign.
The second order came from Minister Ricardo Villas Bôas Cueva, the electoral inspector general, in an investigation brought by Lula's coalition against the Flávio Bolsonaro and Alfredo Gaspar ticket, alleging that platforms favoured the opposing campaign and limited the reach of pro-Lula content. Cueva said the evidence justified opening the case but did not yet prove wrongdoing. He ordered Meta and X to provide technical explanations on AI use, distribution of political content, election ads and removals, with a daily fine of R$500,000 each for unjustified non-compliance. Meta must also report on automatic translations and GIF and hashtag searches, and X must explain how its Grok assistant works.
What it means in practice
The rulings show the court enforcing its synthetic-content labelling rules case by case rather than through blanket measures. Cueva ordered human review of one Instagram removal, a promotional video for an episode of the series Revelando taken down on AI grounds, but rejected a request to require prior human review of every removal of journalism and opinion about the presidential candidates, calling it too broad. Requests to suspend automatic translations and restore the reach of campaign ads will be decided after the platforms reply, and the court stressed that opening the case does not mean the accusations are accepted. The labelling rules were first tested on the eve of the first round, when fact-checkers counted nearly one AI-made electoral post per hour.
What to watch: how Meta and X answer, whether more unlabelled AI material circulates in the final hours before the vote, and whether the investigation leads to sanctions after the election.