← Blog

Before Brazil's run-off: what is documented about influence online

10 October 2026

Brazil chooses its president in a run-off on 25 October. Around the first round, on 4 October, much was said about who might be shaping the campaign online. This post sets out what has been documented, by whom and how, and what has not, so that you can weigh the next claim you meet yourself.

We have kept to official records and to research or fact-checking that publishes its method. We name no candidate and no party in what follows; the titles of some of the sources do. The evidence below concerns practices, and several of them were aimed at both of the leading candidates. Quotations from Portuguese sources are our translation.

The rules

The rules on electoral propaganda of Brazil’s Superior Electoral Court (TSE), amended for 2026 by Resolution 23.755 of 2 March 2026, do four things that matter here:

• They ban deepfakes, audio or video “generated or manipulated digitally” to create, replace or alter a person’s image or voice, used to harm or favour a candidacy.
• Campaign material made with AI must say so, “explicitly, prominently and accessibly”, and name the technology used.
• New for 2026: no publishing, republishing or paid boosting of new AI-made content using the image, voice or words of a candidate or public figure from 72 hours before the end of voting until 24 hours after it, even when labelled.
• Also new: AI systems offered by platforms may not “rank, recommend, suggest or prioritise” candidates, even when asked.

Not everyone agrees with the court’s controls on disinformation. Observers from the Organization of American States (OAS) recorded that some political groups told them disinformation and hateful posts circulated despite them, while others argued that they “limited freedom of expression and constituted a form of censorship”.

What the intelligence agency says, and does not

The public report of ABIN, Brazil’s intelligence agency, Desafios de Inteligência – Edição 2026, published in December 2025, calls the possibility of foreign interference in the election a “risk factor that cannot be underestimated”. It lists the means such interference “can” use: sophisticated disinformation campaigns, cyberattacks on electoral infrastructure, and hidden funding of political groups. It names no country. It is a risk assessment, not a finding. The same chapter describes the platforms’ algorithms as “designed to maximise engagement”.

What has not been found

On the day of the first round, the TSE’s president told journalists: “So far, we have no concrete indication of international interference.” The court’s own account of his closing statement that night reads: “No facts were found that indicated foreign interference in the Brazilian electoral process.” The OAS observers quoted that sentence in their preliminary report of 6 October.

So, as of 9 October, no Brazilian authority has documented a foreign influence operation through social media against this election. A claim that one exists goes further than the record; so does a claim that the question is closed for good.

What has been documented

Paid posts with an unknown payer. The fact-checking outlet Aos Fatos found that at least 11 foreign accounts published posts promoting one presidential candidate on X between 18 and 21 September, carrying X’s own “paid partnership” label. Its report of 25 September states that who paid for them “remains unknown”. Another candidate asked the TSE on 23 September to have the posts removed; we found no ruling as of 9 October.

AI content in volume, aimed at both leading candidates. The VigIA project of the fact-checker Lupa, which uses detection tools from the University of Campinas, collected 1,941 suspect political posts between 16 August and 28 September. Of the 972 it analysed, 920 were confirmed as made or altered with AI. Of those, 554 were deepfakes, and only 371 carried any AI label. The two candidates now in the run-off were the most often depicted, in posts promoting them and posts attacking them. In the window around the first round when new AI content of candidates was banned, 1 to 5 October, Lupa still counted 64 such posts, 37 of them deepfakes.

These are counts, not reach. Agência Brasil reported the TSE’s president as saying on 4 October that the feared spread of deepfakes had not materialised, with no incidents of significant repercussion. Both can be true.

Ads during the ad ban, including ads posing as the court’s. Paid electoral ads are banned online from 48 hours before voting until 24 hours after. NetLab, a research laboratory at the Federal University of Rio de Janeiro, searched the platforms’ ad libraries during that window and found 568 irregular ads on Meta’s platforms. Six more imitated the TSE and offered a non-existent way to vote by phone or from home; some led to sites asking for the visitor’s CPF, the Brazilian taxpayer number. Meta replied that the sample “does not reflect the scale” of its integrity work.

Checking and reporting

• Siade is the TSE’s anonymous channel for false content about voting machines, the voting process, or pages and profiles that imitate the electoral courts. The TSE explains how to use it.
• Pardal is the TSE’s app for irregular campaigning, including unidentified propaganda online and misleading videos of candidates. It asks you to sign in with e-Título or gov.br.
• Fact-checkers that publish their methods: Aos Fatos, Lupa, and Comprova, a project of newsrooms led by Abraji, the Brazilian Association of Investigative Journalism.

A claim that cannot be traced to a document is worth holding loosely, whichever candidate it helps. The wider pattern, and what is known from other countries, is on Elections & Democratic Harm. We will re-read these sources after the run-off, and correct this post if any of them changes.

← All posts
About this project →