When the grading process becomes opaque, the entire system loses credibility.
In Brazil this week, the National Council of Justice suspended a judicial selection exam for Ceará's state court after evidence emerged that artificial intelligence — rather than qualified human examiners — may have graded candidate responses without authorization or disclosure. The case touches something ancient and unresolved: the question of who holds the authority to judge, and whether that authority can be delegated to a machine without the knowledge of those being judged. At stake is not merely one exam, but the credibility of the mechanisms societies use to decide who earns the power to interpret their laws.
- Brazil's CNJ halted the TJ/CE judicial exam mid-process after investigators uncovered signs that an algorithm — not human legal professionals — had been scoring candidate answers.
- Candidates who sat the exam now exist in a limbo of uncertainty, unsure whether their results reflect fair evaluation or the outputs of an unvetted, unauthorized system.
- The deeper disruption is institutional: a gatekeeping process meant to ensure judicial merit has been compromised at its most critical point — the moment of judgment itself.
- The CNJ faces a tangle of costly choices — invalidate the exam entirely, re-grade manually, or chart some hybrid course — each carrying its own burden of time, expense, and eroded trust.
- The case is accelerating pressure on Brazil to establish transparent, democratically deliberated rules for AI use in judicial selection and other high-stakes government processes.
Brazil's National Council of Justice suspended a competitive exam for Ceará's state court system this week after evidence emerged that artificial intelligence may have been used to grade candidate responses — without authorization, disclosure, or proper oversight. The discovery has thrown the legitimacy of the entire selection process into doubt.
Judicial exams are not routine assessments. They are the gatekeeping mechanisms that determine who advances to positions of legal power and public responsibility. When the grading process becomes opaque or potentially compromised, the consequences ripple outward: candidates cannot know if they were evaluated fairly, the court cannot know if it selected qualified judges, and the public loses confidence in the people chosen to interpret its laws.
What makes the case especially troubling is the apparent absence of transparency. There is no indication candidates were informed their answers might be scored by a machine learning model, nor that the CNJ had approved or established safeguards for such a method. The discovery came as a surprise during investigation — not as a declared policy choice.
The human cost is immediate. Those who passed face uncertainty about whether their results will hold. Those who failed must wonder if an algorithm made errors that cost them their careers. The CNJ must now choose between invalidating the exam entirely, conducting a full manual re-grade, or finding another path — each option carrying significant costs.
The incident reflects a wider tension in how institutions are adopting AI — deploying it in sensitive, consequential decisions with minimal public awareness or democratic deliberation. The CNJ's suspension signals that the integrity concern is being taken seriously, but the real measure will come in what follows: whether Brazil establishes clear rules on AI use in judicial selection, whether transparency becomes mandatory, and whether candidates earn the right to know how their work was judged.
Brazil's National Council of Justice halted a competitive examination for Ceará's state court system this week after discovering that artificial intelligence may have been used to grade test answers—a discovery that has thrown the legitimacy of the entire selection process into question.
The CNJ, which oversees the country's judicial system, suspended the TJ/CE exam after investigators found evidence suggesting that an automated system had evaluated candidate responses rather than human examiners. The move affects an unknown number of applicants who took the test expecting their answers would be reviewed by qualified judges or legal professionals. Instead, their work may have been scored by an algorithm no one had authorized or properly vetted.
The suspension raises a fundamental question about who decides who gets to practice law in Brazil's courts. Judicial exams are not casual assessments—they are gatekeeping mechanisms that determine which lawyers advance to positions of significant power and responsibility. When the grading process itself becomes opaque or potentially compromised, the entire system loses credibility. Candidates cannot know if they were evaluated fairly. The court cannot know if it hired qualified judges. The public cannot know if the people interpreting the law were chosen on merit.
What makes this case particularly troubling is not just that AI was used, but that it appears to have been used without proper disclosure or oversight. There is no indication that candidates were told their answers might be evaluated by machine learning models. There is no evidence that the CNJ had approved this method in advance or established safeguards to ensure accuracy. The discovery came as a surprise—a red flag raised during investigation, not a transparent policy choice.
The suspension will have immediate consequences for everyone involved. Candidates who passed the exam now face uncertainty about whether their results are valid. Those who failed may wonder if they were rejected by a system that made errors. The court itself must now decide what to do: invalidate the entire exam and start over, conduct manual re-grading of all answers, or find some other path forward. Each option carries its own costs and complications.
This incident also points to a broader tension in how governments and institutions are adopting artificial intelligence. The technology is being deployed in sensitive areas—hiring decisions, benefit determinations, criminal sentencing recommendations—often with minimal public awareness or democratic deliberation. When things go wrong, as they have in Ceará, the damage is already done. Careers have been affected. Trust has been broken.
The CNJ's decision to suspend the exam suggests the council takes the integrity concern seriously. But the real test will come in what happens next. Will Brazil establish clear rules about when and how AI can be used in judicial selection? Will there be transparency requirements? Will candidates have the right to know how their answers were evaluated? The answers to these questions will shape not just this exam, but how the country approaches automation in other critical institutional processes.
A Conversa do Hearth Outra perspectiva sobre a história
Why does it matter that an AI system graded these exams instead of a human?
Because the people taking the test had no idea. They prepared for a human evaluation, with all the subjectivity and appeal processes that implies. An algorithm operates by rules no one can fully explain or challenge.
But couldn't an AI grade consistently and fairly?
Maybe. But consistency isn't the same as legitimacy. If candidates don't know they're being evaluated by a machine, and if no one has validated that the machine actually works, then the fairness is theoretical. It's not real.
What happens to the people who already passed or failed?
That's the hard part. Their careers may depend on this exam. Some are probably celebrating a pass they might not have earned. Others are devastated by a failure that might have been a mistake. Now everyone is in limbo.
Could they just re-grade everything by hand?
Theoretically, yes. But that's expensive, time-consuming, and it still doesn't solve the trust problem. People will always wonder if the second grading was done differently than the first.
Is this just a Ceará problem, or is it bigger?
It's a symptom. Governments everywhere are quietly putting AI into hiring, benefits, sentencing—places where it directly affects people's lives. Most people don't know it's happening. Ceará just got caught.