AI Regulation Tracker / Regulation in force
Brazil bans election deepfakes and orders "made-with-AI" labels for the 2026 vote
Brazil's top electoral court has codified an AI-disclosure duty, a ban on specified election deepfakes, and a synthetic-content blackout around polling day. Any firm running or hosting Brazilian campaign content now works under an enforceable AI rulebook.
Brazil's Tribunal Superior Eleitoral has amended its existing rules on artificial intelligence in election campaigns for the 2026 vote. Resolução TSE nº 23.755, issued March 2, 2026, amends the electoral-propaganda regulation, Resolução TSE nº 23.610/2019, and sets the terms under which AI-generated media may appear in the October 2026 vote. The instrument does three things at once: it requires disclosure, it prohibits a category of content outright, and it imposes a timed blackout on synthetic media around polling day.
Update, 2 October 2026
The pre-vote window in Article 9º-B, § 3º-A of Resolução TSE nº 23.610/2019, as amended by Resolução nº 23.755/2026, began on Thursday 1 October 2026, on the Tribunal Superior Eleitoral's own account. The first round of the 2026 general elections is on Sunday 4 October.
In a news item dated 01/10/2026 the TSE wrote that from this Thursday "ficam vedadas, até 24 horas após o término da votação, a publicação e a republicação, ainda que gratuitas, e o impulsionamento pago de novos conteúdos sintéticos produzidos ou alterados por inteligência artificial (IA) ou tecnologias equivalentes que utilizem imagem, voz ou manifestação de candidata, candidato ou pessoa pública" (our translation: until 24 hours after voting ends, the publication and republication, even free of charge, and the paid boosting of new synthetic content produced or altered by artificial intelligence (AI) or equivalent technologies that use the image, voice or expression of a candidate or public person are prohibited).
The resolution applies the ban to such content "mesmo que rotulados e em conformidade com as demais exigências deste artigo" (our translation: even if labelled and in compliance with the other requirements of this article). Under § 4º, breach of the caput and of §§ 3º and 3º-A "impõe a imediata remoção do conteúdo ou indisponibilidade do serviço de comunicação" (our translation: requires the immediate removal of the content or the unavailability of the communication service), on the application provider's initiative or by court order.
The prohibition reaches republication as well as publication and paid boosting of "novos conteúdos sintéticos" (our translation: new synthetic content). This desk has not established how the window applies to republication of material first published before 1 October; the texts read here do not define "novos" closely enough to settle it. Where a second round is held, the TSE news item gives 25 October as the voting date. It does not state a separate window for that round, and this desk does not compute one.
The labeling duty
The core obligation sits in Article 9º-B. Under the resolution, propaganda that uses multimedia synthetic content generated by AI carries a duty for the person responsible for the propaganda to inform, in the resolution's words, "de modo explícito, destacado e acessível" (in an explicit, prominent, and accessible manner) that the content was fabricated or manipulated. The rule is specific about placement: the notice must appear on each page or face of printed material where the synthetic content is used. This is not a buried footnote standard. The disclosure has to be visible where the content is.
The reach of the duty turns on two triggers: content that was created by AI, and content that was significantly altered by it. A campaign that uses a generative tool to produce an ad, or to reshape existing footage, falls inside the rule. That covers a wider slice of modern campaign production than most teams assume, since AI now touches image editing, voice work, and video assembly that once counted as ordinary post-production. The practical result is that a labeling decision has to be made at the point of creation, not left to a lawyer at the end of the pipeline.
The deepfake ban and the blackout
Article 9-C, paragraph 1, prohibits using synthetic audio, video or both to harm or favor a candidacy when the content digitally creates, replaces or alters the image or voice of a living, deceased or fictitious person. Authorization does not remove that prohibition. The separate AI-labeling requirements apply under Article 9-B.
On top of that, Article 9º-B, § 3º-A creates a window ban. It prohibits publishing, republishing, and paid boosting of new synthetic content produced or altered by AI that uses the image, voice, or expression of a candidate or public figure from 72 hours before the vote through 24 hours after it. The window is designed to cut off last-minute synthetic media at the moment when there is no time to correct the record.
Platforms and the burden of proof
The resolution reaches beyond campaigns. Article 125-B requires internet service providers to develop conformity plans that address the AI rules, with measurable indicators and implementation timelines. Platforms are expected to show their work, not simply react to complaints. For a service provider, that is a shift from ad hoc moderation toward a documented compliance program with stated targets and dates.
Article 9º-I adds an evidentiary tool. It allows a judge to shift the burden of proof in synthetic-content disputes, requiring the responsible party to demonstrate how and at which stages AI was used and to verify the accuracy of the information. That reverses the usual posture: the party that deployed the content has to account for it.
What it does not do
Article 9-B, paragraph 4, requires immediate removal of content or unavailability of the communication service for the violations it lists, without prejudice to the proceedings described in Article 9-C, paragraph 2. Article 30 separately addresses internet campaigning and prohibits anonymity. For violations of that article, it sets a R$5,000 to R$30,000 fine for the responsible disseminator and a beneficiary whose prior knowledge is proved. Article 30, paragraph 1-A, expressly excludes the internet application provider from that particular fine. The rule also stops short of banning AI in campaigns generally; disclosed, non-deepfake synthetic content remains permitted outside the blackout window. And enforcement is still early. As of April 2026, the TSE had received 141 complaints, 9 of which involved AI use, a sign that the machinery is running but the AI-specific caseload is small so far.
For a US reader, the cross-border angle is direct. There is no US federal deepfake-election law, and state rules are a patchwork. Brazil's TSE now offers the clearest single national template for AI-content disclosure paired with a pre-vote synthetic-media blackout. US firms that build campaign tools, run ad platforms, or advise candidates operating in Brazil are bound by it, and anyone drafting US rules has a working model to study.
Frequently Asked Questions
What did Resolução TSE nº 23.755 actually change?
It amended Brazil's electoral-propaganda rule (Res. 23.610/2019) for the October 2026 elections. Among other changes it bars publishing, republishing or paid boosting of new AI synthetic content that uses the image, voice or expression of a candidate or public person from 72 hours before to 24 hours after the vote (Art. 9º-B, § 3º-A), even if labelled. The AI-use labeling duty in the Art. 9º-B caput predates it: the TSE's compiled text marks that caput as included by Resolução nº 23.732/2024. The TSE's compiled text marks the deepfake prohibition in Article 9-C, paragraph 1, as included by Resolution 23.732/2024.
Who has to comply?
Candidates, political parties, campaign vendors and agencies, and internet platforms that host or boost Brazilian political content. Platforms must also file conformity plans under Art. 125-B, and a judge can shift the burden of proof onto the party responsible for the content under Art. 9º-I.
Does taking content down avoid a penalty?
Removal does not end every possible proceeding. Article 9-B, paragraph 4, requires removal or service unavailability without prejudice to proceedings under Article 9-C, paragraph 2. Article 30 separately addresses internet campaigning and prohibits anonymity. For violations of that article, its R$5,000 to R$30,000 fine applies to the responsible disseminator and a beneficiary with proved prior knowledge. Paragraph 1-A excludes internet application providers from that particular fine.
Does this ban all AI in Brazilian campaigns?
No. Article 9-C, paragraph 1, prohibits synthetic audio, video or both that digitally creates, replaces or alters the image or voice of a living, deceased or fictitious person to harm or favor a candidacy, even with authorization. The separate vote-window rule covers publication, republication and paid boosting of new synthetic content using a candidate's or public person's image, voice or expression, even if labelled, from 72 hours before voting through 24 hours after it. Other AI content remains subject to the applicable election rules.
Sponsored Training
Practical AI training for regulated professionals, built around verification, documentation and a defensible process. See the courses.
Browse the full AI Regulation News tracker
Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.