Brazilian Court Orders X to Delete AI-Generated Posts Defaming Lawyer

The Order That Held X Responsible for Grok’s Output

A judge in São Carlos, São Paulo, has ordered X to delete posts generated by its artificial intelligence chatbot Grok that falsely accused a prominent lawyer of professional misconduct. In an urgent decision, Judge Jorge Passerini of the 5th Civil Court applied a “relative presumption of fault” to the platform because the content was produced by an “artificial dissemination mechanism” maintained by X itself—not by a third-party user.

The posts targeted João de Senzi, a lawyer specializing in digital law and influencer defense. After a user asked Grok about Senzi’s “judicial defeats,” the chatbot claimed he had lost a case because of a “weak strategy” and had submitted a legal petition containing 43 fake precedents allegedly invented by ChatGPT. The AI also stated that Senzi had been fined for bad-faith litigation and reported to the Brazilian Bar Association, none of which actually occurred. Grok cited as sources articles from legal news sites, but Senzi demonstrated in his filing that none of those articles referred to him.

Senzi sued at the end of March, arguing that the posts severely harmed his professional reputation and that X “makes available, promotes and economically benefits from the AI tool without exercising due control over the content it generates.” He seeks damages of R$20,000, though the merits of the case have not yet been judged. Reports indicate that X is now attempting to reach a settlement with Senzi, likely to avoid creating an unfavorable legal precedent.

The case draws on broader concerns about Grok, which has previously drawn scrutiny from Brazilian authorities for generating sexualized images of minors and non-consensual adult images. Federal prosecutors, the data protection authority, and the consumer defense secretariat have already demanded monthly reports from X on measures taken to block such content.

Why This Ruling Could Redefine AI Platform Liability

A Shift in Liability: From Third-Party Content to AI-Generated Posts

The decision marks a significant departure from the usual application of Brazil’s Internet Civil Rights Framework (Marco Civil da Internet). Under Article 19, platforms are generally not liable for user-generated content unless they fail to remove it after a court order. But Judge Passerini distinguished this case by treating Grok’s output not as third-party content but as something the platform itself produces. Legal expert Andressa Barros, CEO of Fragata e Antunes Advogados, explained that this is the first time a judge has held a platform directly responsible for content generated by its own AI tool, bypassing the Supreme Court’s interpretation of intermediary liability for third-party content. The judge essentially said: “You are responsible for the content your platform produces; it’s internal, not external.”

Why This Ruling Could Set a Precedent for All AI Chatbots

If upheld, the reasoning could apply to any platform that deploys a large language model and allows it to generate statements about identifiable individuals. ChatGPT, Gemini, and other AI tools could face similar liability for defamatory or otherwise harmful text they produce, even if the prompt came from a user. The risk is not limited to social networks: any company using customer-facing chatbots that inadvertently defame someone could be exposed. Barros worries that the precedent may be extended carelessly, potentially leading to an environment where even professionals who inadvertently cite AI-invented case law—like the lawyer did with his petition—could face judicial sanctions.

Legal Context: A Pivotal Moment for AI Regulation in Brazil

The case arrives as Brazil’s Congress debates a comprehensive AI regulation bill, the Marco da IA. The Senate passed the bill in December 2024, but it still awaits final approval in the Chamber of Deputies. Key provisions include risk classification for AI systems and a ban on applications that could significantly and irreparably harm people’s physical or moral integrity. This court order, which implicitly treats generative AI as having an inherent risk of causing harm, may accelerate legislative momentum and influence how judiciary bodies interpret the eventual law.

The Broader Crackdown on Grok’s Content Moderation Failures

Grok had already attracted regulatory attention for facilitating the creation of deepfake and sexualized imagery. In February, the Federal Public Ministry, the National Data Protection Authority, and the National Consumer Secretariat demanded that X provide regular reports on steps to curb such content. The defamation ruling adds another layer: it signals that Brazilian authorities are willing to act against platforms whose AI tools generate not just illegal imagery but also false and damaging text. If X settles the case, it may prevent an appellate court from setting an even stronger precedent, but the industry will still take note of the first-instance logic.

What Tech Platforms and Legal Professionals Need to Do Now

For platforms and developers of generative AI tools:

• Conduct an immediate audit of any chatbot’s output regarding identifiable individuals. The São Carlos ruling demonstrates that a platform can be held liable not only for user-generated defamation but for content its own AI fabricates.
• Implement real-time filtering that blocks defamatory, factually unsupported claims about named persons, especially when the AI cites fictitious or unrelated sources. The judge’s decision explicitly cited the chatbot’s references to “fake sources” as a factor.
• Given X’s reported attempt to settle the case to avoid an unfavorable precedent, consider proactively resolving similar disputes where legal exposure is clear. A settlement now may be cheaper than a binding appellate ruling later.

For legal professionals and other practitioners:

• If you use AI tools for research or drafting, verify every claim, case citation, and source the tool provides. The Senzi case exemplifies how a chatbot can invent juridical references and attribute them to a real person, exposing you to disciplinary action.
• Document all AI-assisted work steps, including prompts and outputs, to defend against accusations of bad faith if AI-generated errors slip through.

For policymakers and compliance teams:

• Track the Marco da IA legislation closely. The bill’s risk-classification regime, if enacted, may directly address the type of AI behavior seen here, potentially imposing mandatory safeguards on systems that can affect reputation and personal integrity.
• In the interim, Brazilian authorities have already demonstrated a willingness to use existing frameworks—such as the Marco Civil and consumer protection laws—to demand transparency and corrective action from AI operators. Prepare for similar requests from the ANPD or MPF.