Brazil’s Superior Electoral Court (TSE) has tightened the rules governing political speech on the internet, introducing a series of restrictions on the use of artificial intelligence in election campaigns. The regulations, outlined in a Folha de S.Paulo explainer, aim to curb disinformation, protect the integrity of the vote and set clear limits for voters, candidates, platforms and AI developers.
Voter Expression Limits
The TSE affirms that voters may express their opinions online, but they must be identifiable—anonymous posts are not permitted. If a voter’s message "offends the honour or image" of a candidate, party, federation or coalition, or spreads "known false facts," the expression may be limited. Violations can lead to fines or, in serious cases, the cancellation of candidacies or elected mandates.
Paid Boosting and Transparency Library
Only candidates and parties may pay to boost the reach of election‑related posts on social networks that offer this service, and the boost must benefit the payer, never be used to attack an opponent. Irregular boosting carries fines ranging from R$5,000 to R$30,000, depending on the amount spent, and a candidate who benefited from the boost with prior knowledge can also be penalised.
In 2024 the TSE required platforms to host a transparency library for paid political ads. By 2026 only Meta (owner of Facebook and Instagram) and Kwai have complied, narrowing the field of companies allowed to offer paid boosting services.
Disinformation and AI Content Rules
The resolution expressly forbids using propaganda that contains "fabricated or manipulated" content designed to spread demonstrably false or decontextualised facts that could disturb the electoral balance. Since 2022 the TSE has applied jurisprudence that imposes fines of R$5,000 to R$30,000 for such offences, including against supporters; grave cases can lead to the cassation of an elected candidate or a declaration of ineligibility. The first conviction of this type occurred in 2021.
Electoral Crimes and Gender‑Based Political Violence
Certain conducts may constitute electoral crimes, such as slander, libel or the dissemination of false information. Most of these are considered low‑offense crimes eligible for a simplified procedural form, except in cases of recidivism.
The 2026 election will be the third held under the law against gender‑based political violence. Acts that seek to hinder, obstruct or restrict the political rights of female candidates—including harassment, intimidation, humiliation, persecution or threats—are punishable by one to four years in prison, with increased penalties when the offence occurs online.
Deepfakes, Deepnudes and Chatbots
Using deepfakes—defined as AI‑generated audio or video that creates, replaces or alters the image or voice of a living, dead or fictitious person—is prohibited in election propaganda. The TSE has yet to clarify whether satirical uses are exempt; a pending ministerial meeting may decide whether the ban applies only when there is potential to deceive voters.
The court also bars AI platforms from allowing the creation or editing of sexual or nude imagery involving candidates or candidates’ partners.
Campaigns may not employ chatbots, avatars or any AI system to simulate dialogue with voters, pretending the conversation is with the candidate or another real person.
AI Near Election Day and Labeling
From 72 hours before voting until 24 hours after, the publication, republication or boosting of AI‑generated content that uses a candidate’s image, voice or public manifestation is forbidden. A narrow interpretation of the deepfake rule that limits the ban to content capable of deceiving voters would already cover this period.
When election propaganda regularly incorporates AI‑generated material, the use of the technology must be disclosed. Platforms are expected to provide fields where users can self‑declare AI assistance; some networks already apply such labels automatically.
Restrictions on AI Systems and Big‑Tech Obligations
AI systems such as ChatGPT, Claude and Gemini are prohibited from ranking, recommending, suggesting or prioritising any candidate, campaign or party. They may not issue opinions, indicate electoral preferences, recommend votes or engage in any form of political favouritism or disfavour.
The TSE also imposes several duties on major platforms. They must submit a compliance plan detailing how they will meet electoral rules and, later, a report on its implementation. Additionally, they are required to proactively moderate content related to antidemocratic acts, disinformation about the electoral system, grave threats against electoral‑justice members, hate speech, racism, homophobia and gender‑based political violence, and to offer users appeal channels for moderation decisions.
All provisions are backed by the threat of fines, candidate cancellations or, in the gravest instances, criminal prosecution. The rules reflect the TSE’s effort to adapt electoral law to the rapid evolution of AI‑driven communication while preserving the secrecy of the vote and the fairness of the contest.