Justice André Mendonça, vice president of Brazil's Superior Electoral Court (TSE), proposed on Tuesday that the court adopt a binding legal standard to define which uses of artificial intelligence in campaign material count as an illegal 'deepfake' versus other permitted AI content. Under his proposal, only synthetic audio or video with a 'degree of realism or verisimilitude objectively capable of producing a false perception of authenticity', meaning content convincing enough to pass as a genuine recording, would qualify as a deepfake.
Mendonça made the proposal in an individual ruling on a request from the campaign of presidential candidate Flávio Bolsonaro (PL), ordering the removal of an AI-generated video. According to newspaper O Globo, the video was posted on August 11 by the Instagram account '@forabolsonaronacional' and shows a voice actor speaking in the first person as if he were former banker Daniel Vorcaro, who is in pretrial detention as part of the Compliance Zero investigation, falsely linking the senator to a corrupt kickback scheme known in Brazil as 'rachadinha'. Mendonça classified the video as a deepfake in a preliminary assessment and ordered it taken down and not reposted.
Per O Globo, the proposed standard splits AI-generated election content into two categories already outlined in a TSE resolution passed earlier this year: synthetic material that can circulate as long as it is properly labeled as AI-generated, and deepfakes, which face an outright ban. Mendonça argued that simply labeling content as AI-made 'does not have the capacity to make lawful' material that otherwise meets the criteria for a deepfake. He also said it is not necessary to prove any voter was actually fooled: the objective test is whether the content, given its realism and context, is capable of passing as an authentic record.
Aiming for a uniform standard
Mendonça said he brought the proposal given the 'relevance of the matter' and the expectation that similar disputes will multiply throughout Brazil's 2026 campaign, arguing the court needs more legal certainty and predictability. The proposal, issued as an individual ruling, still needs to go before the full TSE bench. O Globo reported it is expected to be reviewed first through the court's virtual session rather than Thursday's in-person hearing, though any justice could request it be moved to a live session.
Mendonça's proposed standard runs counter to an interpretation that had been gaining ground among other TSE justices. O Globo reported that recent rulings in the court's virtual sessions, led by Justice Estela Aranha, had taken a broader reading of the resolution, treating any AI-generated depiction of a candidate saying or doing something that never happened as improper, regardless of how realistic it looked. Mendonça was on the losing side of those earlier votes before the court paused its review. The Folha de S.Paulo column Painel reported that Flávio Bolsonaro's own campaign already expects to lose the broader case over deepfakes and anticipates the TSE will ban ads using the technology, a reminder that Brazil's top electoral court still hasn't settled a common standard for AI in politics ahead of the 2026 vote.