The safer choice is always deletion.
Em 2025, o Supremo Tribunal Federal do Brasil reescreveu silenciosamente as regras que governam a responsabilidade das plataformas digitais, substituindo uma arquitetura jurídica construída com deliberação pública por um sistema de quatro camadas nascido de uma decisão judicial. O que era clareza tornou-se labirinto: o mesmo conteúdo pode agora gerar consequências legais distintas dependendo de sua natureza, alcance e origem. Na história das democracias, poucos riscos são tão sutis quanto aquele em que a proteção contra o dano se converte, ela mesma, em instrumento de silêncio.
- O STF declarou o artigo 19 do Marco Civil parcialmente inconstitucional, derrubando a exigência de ordem judicial para a remoção de conteúdo e abrindo caminho para um sistema de 'notificação e retirada' como regra geral.
- Quatro regimes de responsabilidade agora coexistem — cada um com gatilhos, ônus probatórios e consequências distintos —, tornando impossível para plataformas e usuários antecipar com segurança o que é permitido.
- O incentivo estrutural criado é perverso: diante da incerteza legal, plataformas tendem a remover conteúdo preventivamente, transformando o medo de responsabilização em censura privada silenciosa.
- Com as eleições de 2026 se aproximando e o PL 215/2015 tramitando no Congresso — que endureceria as penas para difamação online —, o ambiente para o discurso político legítimo se estreita de forma preocupante.
Até 2025, o Marco Civil da Internet operava sobre um princípio claro: plataformas só respondiam por conteúdo de usuários se descumprissem uma ordem judicial de remoção. Havia um juiz no caminho. Havia devido processo. Em junho deste ano, o STF declarou esse sistema parcialmente inconstitucional. Em novembro, o novo marco estava oficializado — e a clareza havia sido trocada por complexidade.
O novo sistema tem quatro camadas. A primeira é o 'notice and take down': uma notificação direta à plataforma, sem necessidade de ordem judicial, já é suficiente para gerar responsabilidade em caso de inércia. A segunda trata especificamente de crimes contra a honra — difamação, injúria, calúnia — com uma lógica híbrida: na primeira postagem, ainda se exige decisão judicial; mas em reincidências ou republicações, basta a notificação da vítima. A terceira inverte o ônus da prova para conteúdo patrocinado, disseminado por bots ou vinculado a falhas sistêmicas graves: a plataforma é presumida culpada e precisa provar sua diligência. A quarta equipara marketplaces a fornecedores diretos sob o Código de Defesa do Consumidor.
O resultado é que o mesmo conteúdo pode gerar consequências jurídicas completamente diferentes dependendo de como foi publicado, por quem, quantas vezes e se houve dinheiro envolvido. Para um gestor de moderação, não há uma regra — há quatro, cada uma com lógica própria. O incentivo óbvio é remover primeiro e perguntar depois.
A ironia é considerável. O Marco Civil original nasceu de anos de debate público amplo, envolvendo especialistas, sociedade civil e cidadãos comuns. A reinterpretação do STF desfez esse equilíbrio sem processo deliberativo equivalente. E o momento é delicado: o Brasil se aproxima das eleições de 2026 enquanto o Congresso avança com o PL 215/2015, que endureceria as penas para difamação online. Somar punições mais severas a um sistema já fragmentado, sem debate adequado, é uma fórmula para a autocensura.
A incerteza jurídica não é abstrata — ela molda o que se diz, o que se ouve e o que desaparece. Quando plataformas não conseguem prever as regras, moderam com mais agressividade. Quando a moderação se torna agressiva, o efeito inibidor se alastra. O crítico político, o jornalista investigativo, o ativista organizando um protesto — todos passam a operar numa névoa. O STF quis proteger pessoas de conteúdo nocivo. Pode ter criado, sem querer, as condições para um dano diferente: o desaparecimento silencioso de vozes que deveriam ter sido protegidas.
Brazil's Supreme Court has fundamentally rewritten the rules governing how online platforms handle content, and the result is a legal landscape so fragmented that nobody—not the platforms, not the users, not the courts themselves—can predict what comes next.
Until this year, the Marco Civil da Internet, Brazil's landmark 2014 internet law, operated on a clean principle: platforms were liable for user-generated content only if they ignored a court order to remove it. That meant a judge had to get involved. That meant due process. In June 2025, the Supreme Court declared that system partially unconstitutional. By November, the new framework was official. The result is a four-tiered liability system that has replaced clarity with complexity.
The first tier is "notice and take down." A person complains directly to a platform—no court order required—and if the platform doesn't remove the content, it can be held legally responsible. This used to apply only to non-consensual intimate images. Now it's the default for most illegal content. The second tier applies specifically to defamation, insult, and slander. Here the Supreme Court created a hybrid: the first time someone posts such content, you still need a court order for removal. But if the same person posts similar content again, or if someone republishes the original, the platform now has an active duty to take it down based on a victim's notification alone. No judge. No second opinion.
The third tier inverts the burden of proof entirely. For sponsored or promoted content, for material spread by automated bots, and for serious crimes linked to systemic platform failures, platforms are presumed liable unless they can prove they acted with reasonable diligence and speed. They must demonstrate innocence rather than have guilt proven against them. The fourth tier treats commercial marketplaces as consumer-facing businesses under Brazil's consumer protection code, making platforms responsible for transactions they merely facilitate, regardless of their size.
What this creates is a system where the same piece of content can trigger different legal consequences depending on its category, its reach, whether it's been posted before, and whether money changed hands. A platform executive trying to set moderation policy faces not one rule but four, each with different triggers and different consequences. The incentive, inevitably, is to remove first and ask questions later. Why risk liability under any of these regimes when deletion is free and immediate?
The irony is sharp. The original Marco Civil da Internet was built through years of public debate involving experts, civil society groups, and ordinary Brazilians. It was designed to balance platform responsibility with free expression. The Supreme Court's reinterpretation has dismantled that balance without the same deliberative process. And the timing is precarious. Brazil is heading toward elections in 2026, and Congress is simultaneously pushing Bill 215/2015, which would increase criminal penalties for online defamation. Adding harsher punishments to an already Byzantine liability system, without proper debate about the consequences for speech, is a recipe for self-censorship.
The legal uncertainty is not abstract. It shapes what gets said, what gets heard, and what gets silenced. When platforms cannot predict the rules, they become more aggressive moderators. When moderation becomes aggressive, the chilling effect spreads. The person who might have posted a sharp political critique, the journalist investigating a public figure, the activist organizing a protest—all of them now operate in a fog. The Supreme Court meant to protect people from harmful content. Instead, it may have created the conditions for a different kind of harm: the quiet disappearance of speech that should have been protected.
Notable Quotes
For the restoration of legal certainty, it is vital to recover what the Marco Civil da Internet had already established—a law that benefited from years of broad public debate involving experts and civil society.— Instituto Sivis (editorial position)
The Hearth Conversation Another angle on the story
Why did the Supreme Court feel it needed to change the original system? What problem were they trying to solve?
The original rule required a court order before platforms had to act. The Court saw that as too slow, too protective of platforms. They wanted faster removal of harmful content—especially defamation and intimate images. But they didn't think through what happens when you remove the judicial check.
So they replaced one rule with four. That seems like it would make things worse, not better.
It does. The original system was simple: platforms had a safe harbor as long as they obeyed court orders. Now there's no safe harbor. Platforms are liable under different rules depending on what the content is. They can't predict their exposure.
What happens to someone who posts something critical of a politician?
It depends. If it's defamatory, the first time it goes up, technically a court order is still required. But the platform doesn't know if it's defamatory—that's a legal question. So they remove it preemptively. If they don't, and someone sues, they're liable. The safer choice is always deletion.
And the person who posted it?
They have no warning. No chance to defend themselves before it's gone. The platform made a business decision, not a legal one. That's the private censorship problem.
Is there a way out of this?
The original Marco Civil was built through real debate. To fix this, Brazil would need to go back to that process—bring in the same experts, the same civil society groups, and actually think through what a four-tier system means for a democracy heading into elections. Right now, there's just urgency and fear.