On 12 October 2026, TSE minister Marco Aurélio Bellizze ordered Flávio Bolsonaro, Carlos Bolsonaro, Pablo Marçal and deputy-elect Rony Gabriel to remove posts that used footage from Chile to attack the pro-Lula rally held on Avenida Paulista on 11 October, according to Congresso em Foco. The deadline was 24 hours from notification. Bellizze called the mix of real events and events that did not happen "extremamente deletéria". The order is preliminary. The people involved can still defend themselves after the Procuradoria-Geral Eleitoral weighs in, and the minister sent the measure to the TSE plenary.
The case is useful because it is a relatively clean one, and the way it is handled will be copied in messier ones.
The strongest case for the order
The argument for acting fast is serious. A runoff compresses the window in which a false image can do damage. A correction that arrives after the vote repairs nothing. The posts were not opinion or satire. They presented footage recorded elsewhere as if it showed a specific domestic event, which is a checkable factual claim. A court that waited for a final judgment would, in practice, decide nothing. The order also looks narrow. According to Congresso em Foco, it covered the authors of the posts that reused the Chilean video. Other named parties, including the PL and deputies Nikolas Ferreira and Lucas Pavanato, were not ordered to remove their posts. The fine Lula's coalition sought, up to R$ 30,000 per post, was not applied.
The platform-facing part is also modest. Meta must make one Instagram post unavailable if its owner does not take it down, a measure subject to review by the TSE plenary. That is one profile, not a general duty to unmask users. It is a targeted order in a named case, which is how takedowns should work.
The legal stack behind it
Three layers sit under this order.
First, the TSE's 2026 election rules, approved on 2 March 2026. Per Agência Brasil, the court unanimously banned posting modified content using the image and voice of candidates or public figures in the 72 hours before the vote and the 24 hours after. It also barred AI providers from recommending candidates, even when asked. It reaffirmed that internet providers can be held liable if they do not remove fake profiles and illegal posts. Aos Fatos adds that unlabeled synthetic content must be taken down immediately and that platforms must file compliance plans covering the periods before, during and after the elections.
Second, the Supreme Court's June 2025 Marco Civil ruling. By 8 votes to 3, the STF declared Article 19 partially unconstitutional (Agência Brasil). Platforms must act after an extrajudicial notice for listed categories, such as terrorism, child abuse material and anti-democratic acts. A court order remains the rule for crimes against honor (calúnia, difamação, injúria). The ruling also describes a general duty of care.
Third, ordinary electoral jurisdiction. The Bellizze order is not a new power. It applies existing electoral rules through a liminar, and the Plenary had upheld a similar precedent on 9 October.
The point worth noticing is that the Chile case needed none of the broader platform duties. It was resolved the old-fashioned way: a judge, a named respondent and a specific post. That is the version of the system that works.
Where the risk sits
The risk is in the machinery being built around such orders. Three concerns stand out.
Preliminary orders are not findings. A liminar removes speech first and hears the author second. Where the facts are plainly wrong, as here, the tradeoff is acceptable. The same tool used on contested claims, such as a statistic, a quote or a satirical edit, is not. Because the order here is not final, we do not yet know whether the Plenary will confirm it, narrow it, or lift it.
"Decontextualization" is a wide category. The hook describes the offence as manipulation by decontextualization. That label fits a video from another country presented as a local rally. Applied loosely, it also fits a cropped clip, a poorly sourced chart or a sharp campaign edit, which is ordinary political speech. Courts should state the factual falsehood in each order. A bare appeal to context is not enough.
Duty-of-care regimes reward over-removal. Once platforms face liability for what they leave up, and little for what they take down, they will take down more. The EFF's October 2026 submission to Brazil's data protection authority makes this point for the new intermediary liability regime: it calls for appeal routes for restricted posts and warns against overreliance on AI-based content moderation. Those safeguards matter most when the removal volume is highest, which is exactly a runoff campaign.
There is also a symmetry problem. This order went against allies of the opposition candidate. The Plenary has also ordered removals of content attacking Flávio Bolsonaro, and the TSE has pushed removals in the other direction in other cases this season. A system that is even-handed in outcomes can still lack legitimacy if every side believes it is being policed more than its rivals.
What proportionate looks like
The pro-innovation, pro-speech position is not that the TSE should do nothing. A mislabeled video of a mass event is a fair target for removal. The position is that the tools should stay narrow:
- Orders should name the specific false claim and the specific posts, as this one did.
- Platform data requests should stay tied to a named profile and a pending case, not become routine surveillance.
- Every preliminary order should go to the Plenary quickly, with the final reasoning published. This one has been referred; the follow-through is what matters.
- Platforms should offer an appeal channel and a record of what was removed and why, so that errors can be reversed.
- The 72-hour synthetic-content window should be applied to deceptive imitation of people, not to all edited political material.
Brazil is running a high-stakes experiment, with a court acting as speech regulator, a Supreme Court ruling shifting platform liability, and election rules that expect platforms to do more. This case shows it can be done narrowly. The test is the case that is not clean, where the facts are disputed and the clock is running.