If the people selecting judges cannot be trusted, the entire edifice becomes questionable.
In Brazil this week, the National Council of Justice suspended a judicial hiring examination for the Ceará State Court, having found evidence that artificial intelligence may have quietly assumed the role of human examiner in grading candidates' answers. The intervention is rare and deliberate — a reminder that the legitimacy of institutions rests not only on their outcomes, but on the trustworthiness of the processes that produce them. At a moment when automation is reshaping public administration worldwide, Brazil's judiciary has drawn a provisional line: efficiency must not be purchased at the cost of accountability.
- Brazil's CNJ halted a state judicial exam mid-process after detecting signs that AI — not qualified human examiners — had been scoring candidates' written responses.
- The suspension throws the careers of exam candidates into uncertainty, forcing a full restart rather than a simple re-grade, signaling the CNJ viewed the breach as too deep to patch.
- The incident exposes a tension running through judiciaries everywhere: automation relieves institutional strain, but in hiring the people who will wield legal power, opacity in evaluation is its own form of injustice.
- Brazil's legal establishment is now confronting the question of whether its cautious embrace of technology inadvertently allowed an algorithm to become a gatekeeper for the bench — without anyone's explicit consent.
- The CNJ's decisive response suggests a precedent is forming: AI involvement in official grading requires explicit authorization and oversight, or the entire process loses its legitimacy.
Brazil's National Council of Justice suspended a competitive examination for the Ceará State Court of Justice this week after evidence emerged that artificial intelligence may have been used to grade candidates' answers. The CNJ, which oversees judicial standards across the country, determined that the integrity of the process had been sufficiently compromised to warrant a full halt rather than a remediation.
These competitive exams are the mechanism through which Brazil selects its judges — designed to be rigorous, transparent, and fair. The stage allegedly affected was answer correction, where human examiners have traditionally been considered irreplaceable: reading responses, weighing them against established rubrics, and assigning scores through a process that is slow but defensible. An algorithm, by contrast, operates according to rules it cannot easily explain, and its decisions are difficult to contest.
The decision to suspend rather than re-grade signals that the CNJ saw no shortcut to restoring trust. Candidates will likely have to sit the exam again, and the court will wait longer to fill its judicial vacancies — a real cost, but one the council appears willing to absorb.
The episode lands amid a broader global reckoning over where AI belongs in official processes. The technology promises to eliminate human bias and reduce costs, but it also introduces new forms of systematic error and opacity. In judicial hiring, the stakes are especially high: a flawed algorithm could quietly disadvantage entire categories of candidates in ways no one notices until the damage is done.
Brazil's CNJ has now made its position legible — AI in grading requires explicit authorization and human oversight, or it has no place in processes that determine who administers justice. Whether this produces a formal policy or remains a singular intervention, the underlying message is unambiguous: institutional legitimacy depends on processes that are not merely efficient, but trustworthy.
Brazil's National Council of Justice halted a competitive examination for the Ceará State Court of Justice this week, citing evidence that artificial intelligence may have been used to grade test answers. The suspension marks a rare public intervention in the mechanics of judicial recruitment and signals deepening anxiety within Brazil's legal establishment about the reliability of automated systems in processes that determine who gets to wear the bench.
The exam in question was administered by the TJCE—the Tribunal de Justiça do Estado do Ceará—as part of its standard hiring process for judicial positions. These competitive exams are the gatekeeping mechanism through which Brazil selects its judges. They are meant to be rigorous, transparent, and above all, fair. The discovery that AI may have been involved in evaluating responses to test questions undermined that basic premise. The CNJ, which oversees judicial conduct and standards across the country, determined the integrity of the process had been compromised enough to warrant suspension.
What makes this intervention significant is not just that it happened, but that it happened at all. Brazil's judicial system has been cautiously experimenting with technological solutions to speed up case processing and reduce backlogs. Automation has seemed like a natural answer to institutional strain. But this incident suggests the judiciary's leadership is not willing to let efficiency override the legitimacy of its own hiring mechanisms. If the people selecting judges cannot be trusted, the entire edifice becomes questionable.
The specifics of how AI was allegedly used remain somewhat opaque from public statements so far. The CNJ identified suspected AI involvement in the answer correction process—the stage where test responses are evaluated and scored. This is precisely where human judgment has traditionally been considered essential. A panel of qualified examiners would read each answer, assess its quality against established rubrics, and assign points. The process is labor-intensive and slow, but it is also defensible. An algorithm, by contrast, operates according to rules it may not be able to explain, and its decisions are harder to contest or appeal.
The decision to suspend the exam rather than simply re-grade it suggests the CNJ found the breach serious enough that remediation was not sufficient. A full restart was necessary. This means candidates who took the test will have to take it again, assuming the exam is eventually reinstated with proper safeguards. It also means the TJCE will have to wait longer to fill whatever judicial vacancies prompted the exam in the first place.
The incident arrives at a moment when courts and government agencies worldwide are grappling with similar questions about where AI belongs in official processes. The technology is powerful and seductive—it promises to remove human bias, speed up decisions, and reduce costs. But it also introduces new forms of opacity and new risks of systematic error. In judicial hiring, the stakes are particularly high. A flawed algorithm could systematically disadvantage certain types of candidates or reward answers in ways that do not actually predict judicial competence.
Brazil's CNJ has now signaled that it will not tolerate AI in grading without explicit authorization and oversight. Whether this leads to a broader policy on AI use in judicial exams, or whether it remains a one-off response to a specific breach, remains to be seen. But the message is clear: the legitimacy of the judicial system depends on processes that are not just efficient, but trustworthy. For now, that means keeping humans in the loop.
Citas Notables
The CNJ determined the integrity of the process had been compromised enough to warrant suspension.— CNJ decision on TJCE exam
La Conversación del Hearth Otra perspectiva de la historia
Why did the CNJ feel compelled to suspend the entire exam rather than just re-grading the answers?
Because if you can't trust the grading process, you can't trust the results. Re-grading might fix the scores, but it doesn't restore confidence in the system itself. The candidates and the public need to know the process was clean from the start.
But surely AI could be used responsibly in grading if it were properly designed and monitored?
Possibly. But this incident shows what happens when it's used without explicit authorization or oversight. The CNJ is essentially saying: not yet, not like this. They're drawing a line.
What's the real cost here—to the court, to the candidates?
The TJCE has unfilled judicial positions that will stay unfilled longer. Candidates who prepared for and took the exam have to do it again. And the court's credibility takes a hit. That's expensive in ways that aren't always visible.
Is this about AI itself, or about transparency?
Both. The core problem is that AI decisions are hard to explain or contest. In hiring, where fairness is supposed to be the whole point, that's a fundamental problem. You can't appeal to an algorithm the way you can appeal to a human examiner.
What happens next?
That's the open question. The CNJ might develop strict guidelines for AI use in judicial exams, or it might decide the risks aren't worth it. Either way, this sets a precedent: automation in judicial hiring is not automatic.