Justice must not only be done, but be seen to be done
In the northeastern Brazilian state of Maranhão, a judge faces twin investigations — one into allegations of sexual harassment, another into suspicion that he used artificial intelligence to generate court sentences riddled with legal errors. The convergence of personal misconduct and technological misuse raises a question older than any algorithm: who is accountable when justice is delegated to a system that answers to no one? The defendants who received these rulings now exist in a kind of legal uncertainty, their fates shaped by reasoning that may not hold under scrutiny. How a society responds to such a failure reveals as much about its institutions as the failure itself.
- A judge in Maranhão now faces two simultaneous investigations — sexual harassment allegations and suspicion of secretly using AI to produce legally defective court rulings.
- Investigators identified sentences bearing the hallmarks of machine-generated text: logical gaps, inconsistent legal reasoning, and conclusions disconnected from evidence or applicable law.
- Defendants who received these rulings may have been denied due process — judged not by an accountable human being, but by an algorithm operating without oversight or disclosure.
- The case forces an uncomfortable question onto the Brazilian judiciary: if this is happening in one courtroom, how many others remain undetected, ungoverned, and undisclosed?
- Authorities now face a choice between treating this as an isolated bad actor or using it as a catalyst to establish binding transparency and oversight standards for AI across all Brazilian courts.
In Maranhão, northeastern Brazil, a judge is under investigation on two fronts: allegations of sexual harassment, and the suspicion that he has been using artificial intelligence to write court sentences — sentences that contain serious legal errors capable of undermining the validity of the cases they decided.
Investigators examining his work found a troubling pattern. The rulings displayed signs of machine-generated text: fractured legal reasoning, logical inconsistencies, and conclusions that failed to follow from the evidence or the law. These were not merely sloppy judgments. They were structurally flawed in ways that could strip defendants of their right to due process — to be judged by a qualified, accountable human being rather than an algorithm trained on no one's behalf.
The use of AI in drafting judicial decisions is not inherently wrong. Done transparently, with proper training, human review, and full disclosure to all parties, it can be a legitimate tool. But a judge deploying AI in secret, without oversight or the knowledge of defendants and their lawyers, violates something foundational: the principle that justice must not only be done, but be seen to be done.
The deeper alarm is the possibility that this is not an isolated case. If one judge in Maranhão has been operating this way, the question of how many others might be doing the same — undetected, unregulated — becomes impossible to ignore. The Brazilian judiciary, like courts worldwide, is navigating the integration of new technology, and this case suggests that governance has not kept pace.
For the defendants whose fates were shaped by these flawed rulings, the road ahead is uncertain. Appeals, retrials, and prolonged legal limbo may follow. What emerges from this moment — whether a quiet dismissal of one bad actor, or a genuine reckoning with how AI is used across the entire system — will say a great deal about the health of Brazilian judicial institutions.
In Maranhão, a state in northeastern Brazil, a judge now faces two separate investigations that together paint a troubling picture of institutional failure. He stands accused of sexual harassment—allegations that alone would warrant serious scrutiny. But layered atop those charges is something perhaps more systemic in its implications: suspicion that he has been using artificial intelligence to generate court sentences, and that those AI-produced rulings contain legal errors significant enough to undermine the validity of the cases they decided.
The allegations emerged as investigators began examining the judge's work product more closely. What they found suggested a pattern: sentences that bore the hallmarks of machine-generated text, with logical gaps, inconsistent legal reasoning, and conclusions that did not follow from the evidence or applicable law. The sentences were not merely poorly written. They appeared to be fundamentally flawed in ways that could affect the rights of defendants who received them.
This is not a minor administrative problem. A sentence generated by an AI system that has not been properly trained on Brazilian law, that has not been reviewed by a qualified jurist, that has not been subjected to the kind of scrutiny a human judge would apply—such a sentence is not merely a sentence. It is a deprivation of due process. It is a defendant being judged not by a person accountable to the law, but by an algorithm accountable to no one.
The case raises urgent questions about how technology is being deployed in Brazilian courts, and whether there are adequate safeguards in place. A judge using AI to draft sentences is not inherently problematic if done transparently, with proper training, with human review, and with full disclosure to the parties involved. But a judge using AI to generate sentences in secret, without oversight, without the knowledge of defendants or their lawyers—that is a different matter entirely. That is a breach of the fundamental principle that justice must not only be done, but be seen to be done.
What makes this case particularly significant is that it may not be isolated. If one judge in Maranhão has been using AI to generate sentences, how many others might be doing the same? Are there systems in place to detect this? Are there protocols requiring disclosure? The Brazilian judiciary, like judiciaries around the world, is grappling with how to integrate new technologies responsibly. This case suggests that integration has outpaced governance.
The defendants whose cases were decided by these flawed sentences are now in legal limbo. Their convictions or judgments may be based on reasoning that would not withstand scrutiny. Appeals may be necessary. Cases may need to be retried. The ripple effects of this judge's conduct will extend far beyond his courtroom.
What happens next will matter. If Brazilian authorities treat this as an isolated incident—a bad actor, a few bad sentences—then the broader problem remains unsolved. But if they use this case as a catalyst to examine AI use across the judiciary, to establish clear rules about when and how AI can be used, to require transparency and human oversight, then something constructive might emerge from this failure. The question is whether the system has the will to do so.
A Conversa do Hearth Outra perspectiva sobre a história
Why does it matter that a judge used AI to write sentences, rather than writing them himself?
Because a sentence is not just words on a page. It is a legal determination that affects a person's freedom, their record, their life. When a judge signs a sentence, he is taking responsibility for it. He is saying: I have read the evidence, I know the law, and this is my judgment. When an AI writes it, that accountability vanishes. No one knows if the AI understood the case. No one knows if it applied the right law.
But couldn't an AI be trained to write good sentences?
Possibly. But that would require enormous care—training on Brazilian law specifically, testing against real cases, human review of every output, transparency about what's happening. There's no evidence any of that occurred here. This appears to be a judge using a tool he didn't understand to do work he didn't do.
What happens to the people whose cases were decided this way?
They're stuck. Their sentences may be legally invalid. They may need new trials. They may have served time based on reasoning that wouldn't hold up in court. The damage is already done.
Is this likely to happen in other courts?
That's the real fear. If one judge figured out how to do this, others might too. And if there's no oversight, no one would know. That's why this case matters beyond just this one courtroom.
What would proper use of AI in courts look like?
Transparency, for one. The parties in a case would need to know an AI was involved. There would need to be human review—a qualified judge reading and approving every sentence. There would need to be training specific to the jurisdiction and the law. And there would need to be accountability if something goes wrong. None of that appears to have happened here.