Hidden instructions designed to alter how AI systems process legal documents
In Porto Velho, Brazil, a judge has imposed what appears to be the country's first judicial sanction for the manipulation of artificial intelligence through hidden commands embedded in a legal petition — invisible to human eyes but legible to machines. A health insurance plaintiff, seeking coverage for reconstructive surgery, had buried directives in white text instructing AI systems to favor their claims. The court treated this not as procedural cunning but as a betrayal of the foundational trust upon which justice depends, reminding us that as institutions adopt new tools, those who would subvert them will follow close behind.
- Hidden inside a routine health insurance lawsuit, someone planted invisible instructions designed to steer AI systems toward a favorable ruling — a deception that only came to light when the opposing party noticed something the human eye could not see.
- The discovery forced a Brazilian court to confront a threat its ethics codes were never written to anticipate: litigants who understand AI well enough to exploit it, and who are willing to do so inside a legal filing.
- The judge refused to treat the technique as a gray area, imposing a financial penalty of ten percent of the case's adjusted value and framing the conduct as a deliberate assault on the integrity of electronic judicial proceedings.
- Beyond the fine, formal disciplinary referrals were sent to bar associations in three states, signaling that the consequences will follow the lawyers who signed the petition — not just the document itself.
- Courts and technology teams are now being asked to build defenses they had not yet imagined needing, as this case makes plain that AI-assisted justice is only as trustworthy as the documents it is asked to read.
A judge in Porto Velho, Rondônia, has sanctioned a plaintiff for embedding hidden commands inside a lawsuit petition — an attempt to manipulate any artificial intelligence system that might analyze the filing. The case, involving a health insurance dispute over reconstructive surgery following gastric bypass, appears to be the first of its kind in Brazil.
The hidden instructions were written in white text on a white background, placed after the lawyers' signatures where no human reader would notice them. They were precise and purposeful: directives telling AI systems to classify the procedures as reparative rather than cosmetic, to treat the insurance denial as abusive, to emphasize urgency, and to presume the plaintiff had suffered moral damages. The technique — known as prompt injection — is designed to alter how machine learning systems process and respond to a document.
When the insurance company flagged the discovery, the judge examined the filing and confirmed the allegations. His written decision made clear that this was not a procedural misstep but a deliberate breach of litigation ethics — one capable of influencing not only the immediate request for emergency relief but also any future appellate or AI-assisted review of the case.
The court imposed a financial penalty equal to ten percent of the case's adjusted value, citing the gravity and technical sophistication of the conduct. It also issued formal notices to bar associations in São Paulo, Paraná, and Rondônia, opening the door to professional discipline for the lawyers who signed the petition, and alerted the tribunal's IT and AI departments to strengthen their defenses.
The case lays bare a vulnerability that courts have been slow to reckon with: as AI tools become embedded in judicial workflows, they become targets for those who understand how to exploit them. The commands used here were not technically complex — they were plain-language instructions — but they were enough to reveal that the rules governing legal ethics were written for a world that did not yet include invisible text and machine readers. How Brazil's courts and bar associations respond will set the terms for what comes next.
A judge in Porto Velho, in Brazil's Rondônia state, has penalized a plaintiff for attempting to manipulate artificial intelligence systems through hidden commands buried inside a lawsuit petition. The case marks what appears to be the first judicial sanction of its kind in Brazil—a direct confrontation between courtroom ethics and the emerging vulnerabilities of AI-assisted legal analysis.
The plaintiff, a health insurance beneficiary seeking coverage for reconstructive surgery following gastric bypass, filed suit against the insurance provider. Somewhere in the drafting process, someone embedded instructions designed to influence any AI system that might analyze the document. These commands were hidden in white text on a white background, placed after the lawyers' signatures where they would be invisible to human readers but potentially detectable by machine learning algorithms scanning the petition.
When the insurance company flagged the discovery to the court, the judge examined the filing and confirmed what had been alleged. The hidden instructions were explicit and numerous: directives telling AI systems to classify the procedures as reparative rather than cosmetic, to recognize the insurance denial as abusive, to emphasize the urgency of the case, and to presume the plaintiff had suffered moral damages. The technique is known as prompt injection—a method of inserting covert instructions into documents to alter how AI systems process and respond to information.
The judge's written decision treated this not as a clever procedural maneuver but as a fundamental breach of litigation ethics. He noted that the commands were designed to influence both the immediate decision on whether to grant emergency relief and the underlying merits of the case itself. Even though the request for emergency relief had already been denied, the judge recognized that these hidden instructions could shape how the case was reviewed on appeal or analyzed in future proceedings. The sophistication of the technique—the deliberate concealment, the precise targeting of AI decision-making criteria—suggested this was not accidental but intentional.
The court imposed a financial penalty equal to ten percent of the case's adjusted value, citing the gravity of the conduct, the technical sophistication involved, and the broader threat such practices pose to the integrity of electronic legal proceedings. But the sanction went beyond money. The judge issued formal notices to the bar associations in São Paulo, Paraná, and Rondônia, initiating potential disciplinary proceedings against the lawyers who signed the petition. He also notified the tribunal's information technology department and its AI center, requesting they strengthen security measures to detect and prevent similar manipulation attempts in the future.
The case exposes a gap that few courts have yet had to confront: as judicial systems increasingly rely on AI tools to manage caseloads, analyze documents, and assist in decision-making, they become vulnerable to actors who understand how these systems work and how to exploit them. The hidden commands in this petition were not sophisticated by AI standards—they were straightforward instructions in plain language. But they revealed that the intersection of legal practice and artificial intelligence creates new opportunities for misconduct that traditional ethics rules were not written to address.
What happens next will likely shape how Brazilian courts approach AI security and legal ethics in the years ahead. The referral to the bar associations suggests this will not be treated as an isolated incident but as a matter requiring professional discipline and institutional response. The involvement of the tribunal's technology center indicates courts are beginning to think systematically about how to defend their AI systems against manipulation. For lawyers and litigants, the message is clear: the courtroom is no longer a space where hidden instructions can be safely embedded, even in text invisible to the human eye.
Notable Quotes
The conduct has potential to compromise the integrity of the decision-making process and the reliability of document analysis— Judge in Porto Velho civil court ruling
The Hearth Conversation Another angle on the story
Why would a lawyer risk their career on something like this? It seems almost reckless.
Because they believed—or their client believed—that the case was worth the gamble. When you're fighting an insurance company over coverage, the stakes feel existential to the person denied surgery. The lawyer may have seen AI as a tool that could be influenced the way any other tool can be, without fully grasping that courts would eventually catch on.
But they hid the commands in white text on white. That's not accidental. That's deliberate concealment.
Exactly. That's what made the judge treat it as bad faith rather than mere incompetence. The deliberateness is what transformed this from a mistake into a violation of the fundamental duty of honesty to the court.
Do you think other lawyers have tried this and gotten away with it?
Almost certainly. This case only surfaced because the insurance company was paying attention and the judge took the allegation seriously. How many petitions are filed every day with hidden text? How many courts are actually looking for it? This case is notable precisely because someone caught it and had the courage to sanction it.
What does this mean for AI in courts going forward?
It means courts have to choose: either they build security into their systems now, or they'll spend years fighting manipulation attempts. The judge understood that. That's why he didn't just fine the plaintiff—he sent notices to the technology center. He was saying: this is a problem you need to solve before it becomes routine.