Is Artificial Intelligence Making Magistrates and Servants of the Brazilian Judicial Power Mentally Ill? ()
1. Introduction
Innovations brought about by artificial intelligence have invariably invaded Brazilian reality. Whether it is unlocking a smartphone or identifying a person at an airport or event through facial recognition; or monitoring security through automatic detection of suspicious activities, vehicle license plates or traffic patterns; or offering optimized routes after automatic analysis of traffic data; or optimizing inventory control and planning by analyzing historical sales data and market trends; there is artificial intelligence acting concretely in everyday life.
And the insertion of artificial intelligence in the world of Law is no different. Although there is no news of any tool that, in Brazil, acts or has been developed by artificial superintelligence (ASI), surpassing human cognitive capacity, the Justice 4.0 panel of the National Council of Justice (CNJ), updated until 02/08/2025, reveals that 62 Courts have 140 AI projects under development, with 43 of these projects already in use [1].
The Court of Justice of the State of Rio Grande do Sul (TJRS) alone has 12 projects under development, including projects to automate procedural documents to prepare a sentence report; summarize documents by generating case summaries; search for similar cases using a Chatbot; and transcribe hearing videos. The Court of Justice of the State of Ceará (TJCE) has 7 projects under development with the aim, among others, of automating procedural documents so that legal documents can be classified, including initial petitions; grouping cases by topic, searching for similar cases, and standardizing legislation and case law for the purposes of analyzing precedents using AI; identifying predatory litigation by implementing the Berna system [2]; and classifying risk for the purposes of analyzing revictimization in domestic violence. The Court of Justice of the State of São Paulo (TJSP) has 6 projects under development through which it seeks to automate procedural documents and search for similar cases to analyze at the expense of preparing appeals; suggest movements for dispatch involving the analysis of appeal preparation costs; classify documents for analysis of precedents; and search for similar cases to identify processes with linked precedents [3].
In the Labor Court, the 7 projects of the Regional Labor Court of the 3rd Region (TRT3, Minas Gerais) aim, among others, to automate procedural documents, search for similar cases, consult legislation, compensate for the portion of digitized documents and standardize legislation and case law for the purpose of serving as a draft assistant; and classify documents, consult legislation to monitor decent work. In turn, the 3 projects of the Regional Labor Court of the 4th Region (TRT4, Rio Grande do Sul) aim to search for similar cases through unsupervised learning (Clustering), in order to identify common patterns and characteristics among the information, facilitating data analysis; and stimulate conciliation through the AI conciliability index. The two projects of the Regional Labor Court of the 9th Region (TRT9, Paraná) seek to implement the search for similar cases through a chat platform and a support tool for preparing judgments, so that artificial intelligence can assist in the search for second-degree judgments. In any case, even though each Court has one or more AI projects for different reasons, the fact is that 41 Courts in the country (43.61%) have started using AI tools with the aim of increasing productivity (volume/time) [4].
However, although the CNJ’s static panel (fed with information from the Judiciary’s national database) revealed that the Regional Labor Court of the 8th Region (TRT8, Pará and Amapá) and the Regional Labor Court of the 13th Region (TRT13, Paraíba) did not have any AI projects under development and/or use, it cannot be forgotten that the TRT of the State of Pará and Amapá and the TRT of the State of Paraíba were awarded, respectively, the 2024 CNJ Quality Award - Excellence, for having met 95.2% of the award requirements, and the 2024 CNJ Quality Award - Diamond, for having met 92.0% of the award requirements, including those related to the productivity axis [5].
The TJSP did not win any CNJ Quality Awards in 2024, but the comparison between the 2021 and 2022 data from the CNJ Statistics panel [6] showed that the São Paulo Court of Justice had judged more cases in 2024 (8,857,772 cases) than in 2023 (5,363,880 cases), in 2022 (5,011,970 cases) and in 2021 (4,590,003 cases), even without having an AI technology project under development and/or use directly aimed at productivity, making us reflect on whether what was motivating 43.61% of the Courts to use Artificial Intelligence (productivity) would justify public spending on this matter, as the statistical data showed that there were bodies of the Judiciary that had improved their performance by means other than AI, not specified in the reports.
The topic becomes even more intriguing when we consider that, in early February 2025, the Brazilian Labor Court incorporated three artificial intelligence tools, including Chat-JT, which automated intelligent queries to internal databases to optimize searches for laws and case law and, consequently, assist in more precise analysis of documents and more accurate strategic decision-making [7]. This artificial intelligence was implemented even though the Labor Court judged many more cases in 2024 (5,482,413 cases) than in 2023 (4,604,971 cases), in 2022 (3,912,835 cases) and in 2021 (3,554,398 cases), without Chat-JT existing until then [6].
The study becomes even more provocative when it is found that none of the AI projects were designed, motivated or even used specifically to prevent judges and civil servants from becoming ill, considering that the Minutes of Ordinary Corrections recorded that anxiety disorder, adjustment disorder and recurrent depressive disorder were among the five most frequent causes of absenteeism. Even though TRT18, TRT22, TRT8 and TRT13, among others, were the winners of the 2023 CNJ Quality Diamond Award, and TRT8 won the 2024 CNJ Quality Excellence Award and TRT3, TRT18 and TRT13, among others, won the 2023 CNJ Quality Diamond Award, leading to the question of whether the AI tools used by the Judiciary were created only to achieve goals imposed by the CNJ; whether the AI tool should not contain some containment mechanism to forcibly impose the disconnection of the AI; and whether the illness of judges and civil servants was a result of the AI tools or of how the managers of the Judiciary were dealing with this new reality, especially in light of Objective 8 of the UN 2030 Agenda [8].
Seeking to answer these questions, without, however, concluding all the issues surrounding the topic under study, quantitative research was used, addressing numerical data and public statistics on the topic, as well as descriptive qualitative research seeking to investigate the phenomenon through supporting literature, review of legislation and judicial analysis, in order to understand whether the AI tools used by the Judiciary Branch were aligned with the fundamental social right to protect workers against automation, as provided by law (art. 7, XXVII, [9]) and with the innovations applied in the area of public management.
In this way, the article will contribute to the academic debate because it deals with an intriguing and relevant topic, the content of which remains current, but also to reflections regarding the need for paradigm shifts, given that the creation of technological innovations does not keep up with the same speed as the needs of workers’ bodies and innovations in terms of personnel management.
2. The Need or Lack Thereof for Technological Innovation in the Judicial Field for the Purpose of Increasing Productivity
In fact, the integration of artificial intelligence into the Brazilian judicial system is not something posthumous, but a present and emerging reality. The Justice 4.0 panel of the National Council of Justice [10] reveals that 62 Courts had 140 AI projects under development. Of these projects, 63 were in production, of which 11 had been finalized and not implemented, 3 had not been started, 17 were in the initial stage and 46 were already in use.
Even though some legal professionals claim to miss leafing through the papers in the case files, a comparison of the Justice in Numbers 2024 report [10] with the Justice in Numbers 2023 report [11], the Justice in Numbers 2022 report [12] and the Justice in Numbers 2021 report [10], all from the CNJ, shows that the number of new cases in the Brazilian State Courts has only increased over the years, rising, respectively, from 2021 (21,058,779 actions) to 2022 (23,947,159 actions), 2023 (25,956,591 actions) and 2024 (26,088,214 actions). In the Federal Court, in turn, although the number of new cases varied from 2021 (4,993,408 actions) to 2022 (4,609,246 actions) and from 2023 (6,017,943 actions) to 2024 (5,350,122 actions), the number of actions was never low enough to cause surprise and impact the daily demand for service, confirming that urgent changes needed to be made so that the Judiciary could carry out its primary and inescapable function of resolving conflicts of interest.
Likewise, in the Brazilian Labor Court, the number of cases grew from 2021 (2,955,597 actions) to 2022 (3,277,326 actions), 2023 (4,220,960 actions) and 2024 (4,736,482), thus ratifying that something needed to be done so that society as a whole was served satisfactorily, especially considering that the number of employees in all branches of the Judiciary, despite having suffered a small reduction in 2022 (266,338 employees, including permanent, seconded, requisitioned and commissioned employees), had been increased from 2021 (267,613 employees, including permanent, seconded, requisitioned and commissioned employees) to 2023 (272,060 employees, including permanent, seconded, requisitioned and commissioned employees). 2024 (275,581 employees, including permanent, seconded, requisitioned and commissioned employees) and 2025 (up to 02/12/2025, when the survey ended—283,957).
The existence of magistrates and civil servants is essential for the provision of jurisdictional protection, with the aim of pacifying legal wars and, consequently, society. However, the comparison between the total expenses of the national Judiciary in 2021 (100.06 billion reais), 2022 (103.9 billion reais), 2023 (116 billion reais) and 2024 (132.8 billion reais), with the cost of justice services per inhabitant in 2021 (R$ 475.51), 2022 (R$ 489.91), 2023 (R$ 540.06) and 2024 (R$ 653.70) and the total collections of the Judiciary in 2021; 2022; 2023; and 2024, undoubtedly proves not only that the math does not add up, but also that the calculation will hardly reach zero, as the number of lawsuits grows vertiginously year after year, expenses only increase (although not at the same speed at which the lawsuits are distributed), even to deal with the number of cases in progress in the judicial units, and the collection of revenues by the Brazilian Judiciary cannot keep up with the growth in expenses [13] [14].
But in this scenario, adding the fact that in 2024 there were 18,265 magistrates in office, as opposed to the 18,117 magistrates in office in 2023, as well as the fact that in 2022 there were 18,035 magistrates in office, as opposed to the 17,988 magistrates in office in 2021, the use of AI projects by the Judiciary, mostly created by the Tensorflow Frameworks (reusable codes written by Google Brain that, among others, allow high-level abstractions, transforming raw data into formats suitable for machine learning models), using the Bert language model (an algorithmic structure designed to understand text in natural language, allowing the development of systems without starting from scratch), the Lightgbm algorithms (a sequence of instructions that uses decision trees to build learning models), Regex (a sequence of characters that forms a pattern to, among others, perform operations of text search and editing) and Xgboost (a sequence of instructions that uses decision trees to produce prediction models that are trained iteratively to correct the errors of previous predictions), as well as the Scikit Learn library (structures that provide modules to, among others, detect which predetermined category an action would belong to) [5] it proves to be adequate and reasonable, whether traditional (designed to perform a specific task, learning by examples, with human supervision, such as reproducing a dispatch that has already been made for a similar procedural issue) or generative (with the capacity to create and produce information, content, data, videos, music or other new art autonomously, based on unsupervised or semi-supervised algorithms).
This is because, in addition to the CNJ’s 2023 Artificial Intelligence Research Panel [3] having revealed, on a scale of 1 to 5, that these AI projects had increased efficiency and agility by 4.76, the accuracy and consistency of repetitive tasks by 4.16, the improvement in decision-making by 3.97 and the reduction of errors by 3.86 and, therefore, contributed to increasing productivity by handling the thousands of cases that are being processed in the Brazilian Justice system, the same projects would contribute to containing the increase in personnel expenses that had jumped from R$92,690,856,920.00 in 2021 to R$119,721,831,157.00 in 2024, since the increase in productivity per judge/employee would make the immediate hiring of new employees unnecessary [5].
On the other hand, despite the parallel between the Magistrate Productivity Index (MPI) and the Civil Servant Productivity Index (CPI) of the 2021 Justice in Numbers Report [13] (reduction, respectively, of 22% and 21.7%) with the MPI and CPI of the 2022 Justice in Numbers Report [12] (increase, respectively, of 11.6% and 13.3%), as well as between the MPI and CPI of the 2023 Justice in Numbers Report [11] (increase, respectively, of 10.7% and 10.5%) with the MPI and CPI of the 2024 Justice in Numbers Report [10] (increase, respectively, of 6.8% and 5%), showing that the productivity of magistrates and civil servants had increased as of 2021, indicating that AI projects were impacting productivity indicators, the same numbers reveal that the increase from 2022 to 2024 did not follow an exponential scale.
This is because, in addition to the reported productivity not following a constant/annual increase and/or linear or gradual growth, the substantial gain in performance in legal tasks by judges and civil servants resulting from the use of techniques and algorithms that allowed systems to recognize more elaborate and more diverse patterns (captured from equally increasing amounts of data) did not guarantee the ability of the AI itself to understand or control its indicative results, much less that the AI could not encounter certain insurmountable limits (imposed by the algorithms and data themselves), with regard to computing and the field of modeling and software engineering (abstract concepts that act according to the basic rules of advancing, retreating, including and excluding and do not allow the AI, for example, to have instinctive behaviors), as the numbers themselves showed.
It may even be that the tools used will no longer advance and the IPM and the IPS will become “rigid”, given their own human limitations. After all, the judicial public service still requires human labor, whether due to the variability of details that make each case unique and, consequently, each action, even if similar, distinct from the other; or because the act of judging presupposes a certain amount of common sense, empathy and analysis of the human factual context so that no pretext not authorized by law or partial interpretation is generated; or because, despite the legal system being full of open/imprecise terms, art. 20 of Decree-Law No. 4,657, of September 4, 1942, itself established that decisions would not be made based on abstract legal values without considering the practical consequences of the decision [15].
It is also clear that, despite the CNJ’s static panel, fed with information from the Judiciary’s national database, revealing that the TRT of the 8th Region (TRT8, Pará and Amapá) and the TRT of the 13th Region (TRT13, Paraíba) did not have any AI projects under development and/or use, both were awarded, respectively, the 2024 CNJ Quality Award - Excellence, for having met 95.2% of the award requirements, and the 2024 CNJ Quality Award - Diamond, for having met 92.0% of the award requirements, including those related to the productivity axis [4].
Likewise, the TJSP did not win any CNJ Quality Awards in 2024, but the comparison between the 2021 and 2022 data from the CNJ Statistics panel proved that the São Paulo Court of Justice had judged more cases in 2024 (8,857,772 cases) than in 2023 (5,363,880 cases), in 2022 (5,011,970 cases) and in 2021 (4,590,003 cases). Despite not having an AI technological project under development and/or use directly aimed at productivity.
This, added to the news recently published on the Uai portal (according to which people’s critical thinking was on its last legs due to the harmful effects that the consumption and rapid production of information, as well as the superficial conclusions of AI, generated [16]), seems to prefigure an apocalypse in which AI projects, in addition to not serving the intended purposes of exponential productivity within the judiciary, would end up hindering the formation of critical thinking on the topic at hand, either due to the shallow information provided or because judges and civil servants would end up adopting (for convenience) the content summarized by AI, losing over time the ability to analyze in depth the merits of controversial issues.
However, neither the lack of exponential growth in the Brazilian IPM and IPS, nor possible insurmountable limits of AI, nor the situation in which some courts have managed to increase their productivity without the intervention of AI, nor the possibility of harm to the critical reflection of judges and civil servants, none of this would lead to the presumption that AI projects should not be included in the Judiciary. After all, in addition to the reports of Justice in Numbers and the static panel of the CNJ having revealed that such projects had indeed contributed to the increase in the productivity of the Judiciary in general, as already explained above, artificial intelligence continued to be a plausible, proportional and tangible solution, given the vertiginous increase in lawsuits, the number of lawyers in Brazil, the unauthorized investment of some lawyers by clients and lawsuits with a high likelihood of conviction, and the continued disrespect for the provisions established by the legal system, including the CLT, by individuals and entrepreneurs.
It also remained proportional because the 2023 artificial intelligence research panel [3] showed that, of all the projects developed, 39 were being produced in partnership with other courts or entities in the legal field; 23 in partnership with universities; 13 in partnership with the private sector; and 3 in partnership with other public entities, such that the expenses were not being fully assumed by the federative entities, in view of the savings that the partnerships created.
Contrary to pointing out the possible alleged unnecessary insertion of AI in the Judiciary, the fact that some courts have increased productivity or won the 2024 CNJ Quality Award [4] Excellence and Diamond, even without any AI project, signals and confirms that AI will hardly be able to reproduce all the characteristics of human beings working in the Judiciary, even though 35.7% of the AI projects in the Brazilian Judiciary contemplate the reuse of raw data to learn to recognize patterns or correlations, in order to carry out more precise interferences based on new inputs.
Precisely because human beings are spontaneous creators, often inspired by their personal history, everyone is capable of having unimaginable ideas and improvising, creating solutions, new knowledge and new experiences, to the point of allowing a given Court to increase productivity even without the use of AI. Furthermore, the ethical capacity to discern between good and evil enables human beings to make moral judgments and act accordingly, transforming their characteristic of acting with intention, given their notion of moral responsibility.
This is why judges and civil servants feel encouraged to act in a certain way, when they see that certain innovative procedures allow citizens to have their cases resolved more quickly, as they know that providing an excellent service is part of their duties. AI, on the other hand, does not have spontaneous actions because it is designed and programmed by a person. It has no ethics and, much less, does it act with intention. “A machine is neither good nor bad. It is efficient. It does what it is told to do and what it was programmed to do” [17].
Thus, considering that the insertion of AI in the Judiciary is more than justified by the Brazilian reality, as well as that AI will hardly be able to copy all the characteristics of human beings who work in the Judiciary, it is high time to realize that artificial intelligence must be used as an instrument that enhances the functional skills of judges and civil servants, and not as a means of disqualifying or stifling their innovative capabilities.
More than that: it is time to understand that the biggest problem is not whether or not AI needs to be inserted into the Judiciary. The problem lies in the way AI is applied and used. That is, in moving towards allowing historical decisions to be made by non-human intelligence, allowing errors to generate potentially catastrophic effects, as mentioned by Yuval Noah Harari [18], due to computational public policies that transform computers, algorithms and AI into historical agents.
3. Is Artificial Intelligence Inserted into the Judiciary Contributing to the Illness of Judges and Civil Servants?
In fact, although the CNJ’s 2023 artificial intelligence research panel [3] did not set the precise date on which the first AI project was introduced in the Brazilian Judiciary, the graph “Stage of Evolution of AI Projects”, accessible through the “Comparative” tab of said panel, indicates that the first tool would have been effectively implemented in 2020, as the control transfigured that, in the year 2020, 33 projects had been completed and were in production; and in the year 2022, 63 projects had been completed and were in production; and, in the year 2023, 63 projects had been completed and were in production.
However, even if the above perception does not prevail and the “new era” was initiated [19] by the implementation of the “Victor” program by the STF, at the end of October 2023 by the then president of the Supreme Court, Minister Rosa Weber, the fact is that the comparison between the Health Panel of Magistrates and Civil Servants of the Judiciary of 2020 [20], 2021 [21], 2022 [22] and 2023 [23] demonstrated that the number of magistrates and civil servants who needed to leave public service due to mental and behavioral disorders only increased from one year to the next, with a small variation upwards or downwards in the percentage between men and women.
In fact, if in 2020 there were 174 magistrates (92 women and 82 men) affected by Other Anxiety Disorders (ICD F.41); 137 magistrates (77 women and 60 men) diagnosed with Reactions to Severe Stress and Adjustment Disorders (ICD F.43); and 111 magistrates with Depressive Episodes (ICD F.32), as well as 6240 employees (4192 women and 2048 men) affected by Other Anxiety Disorders (ICD F.41); and 5179 civil servants diagnosed with Depressive Episodes (ICD F.32) [20]; in 2021 there were already 195 magistrates (119 women and 76 men) detected with Other Anxiety Disorders (ICD F.41); as well as 5709 civil servants (3715 women and 1994 men) diagnosed with Other Anxiety Disorders (ICD F.41); 4969 civil servants (3497 women and 1472 men) affected by Depressive Episodes (ICD F.32); and 2671 civil servants (1734 women and 937 men) identified with Reactions to Severe Stress and Adjustment Disorders (ICF F.43) [21].
Likewise, the exponential scale continued to rise in 2022, with 419 magistrates affected (243 women and 176 men) with Other Anxiety Disorders (ICD F.41); 325 magistrates (181 women and 144 men) diagnosed with Reactions to Severe Stress and Adjustment Disorders (ICD F.43); and 233 magistrates (134 women and 99 men) with Depressive Episodes (ICD F.32), as well as 11,592 civil servants (6817 women and 4775 men) affected with Other Anxiety Disorders (ICD F.41); and 8008 employees (4992 women and 3016 men) identified with Depressive Episodes (ICD F.32) [22].
In 2023, there were already 911 magistrates diagnosed (763 women and 148 men) with Other Anxiety Disorders (ICD F.41); 391 magistrates (307 women and 84 men) diagnosed with Reactions to Severe Stress and Adjustment Disorders (ICD F.43); and 233 magistrates (134 women and 99 men) with Depressive Episodes (ICD F.32), as well as 21,186 civil servants (13,397 women and 7789 men) affected by Other Anxiety Disorders (ICD F.41); and 10,176 civil servants (5945 women and 4231 men) identified with Depressive Episodes (ICD F.32) [23], confirming the alarming increase in mental and behavioral disorders among magistrates and civil servants.
It is known that the disease diagnosed as ICD F.41 [24]—Other Anxiety Disorders, presents symptoms such as, for example, constant and exaggerated worry, tachycardia, sweating and feeling short of breath. Its causes are multifactorial, including genetic predisposition, biological factors and environmental triggers. It is also known that the disease cataloged as ICD F.43 [25]—Reactions to Severe Stress and Adjustment Disorders, causes anxiety, depression, insomnia, mood swings and difficulty concentrating, among others, and arises after traumatic events such as the loss of loved ones, serious financial problems, natural disasters, among others. Likewise, the disease described in ICD F.32 [26]—Depressive Episodes, causes sufficiently severe or persistent sadness, capable of interfering with the individual’s general functioning and decreasing interest or pleasure in activities. The cause may be hereditary, or due to changes in neurotransmitter levels, or even due to some alteration in neuroendocrine function or due to psychosocial factors.
However, the fact that the most diagnosed mental illnesses in the Brazilian Judiciary have varied causes and have increased due to Covid-19, in itself, would not prevent the reasoning that the increase in productivity generated by AI projects, due to the increase in efficiency and agility, precision and consistency of repetitive tasks and the reduction of errors, would be used as one of the most important elements to justify the increase in targets by the CNJ and, consequently, contributing to the increase in mental and behavioral disorders among judges and civil servants, by forgetting that behind the targets there was a human being whose agility and precision could not be compared with that of the AI project.
Indeed, the CNJ report “Diagnosis of the mental health of judges and civil servants in the context of the Covid-19 pandemic” from 2022 revealed that only 2% of judges and civil servants had tested positive for Covid-19 in 2020, just as 22.9% of judges and civil servants had tested positive for Covid-19 in 2021 [27], clearly demonstrating that Covid-19 was not the primary cause of the growth of the aforementioned disorders.
In turn, the comparison between the period shown in the graph “Stage of Evolution of AI Projects” (from 2020 to 2023), with the period of insertion of the “Victor” program by the STF, set by Cleylton Mendes (end of October 2023) [19], and the data from the latest survey carried out by the CNJ on mental illness among judges and civil servants [27], pointed to the vertiginous growth of these mental and behavioral disorders mentioned, at the same time that productivity increased and the goals set by the CNJ were raised and/or kept high.
It can be seen that, despite the national percentage of absenteeism of magistrates and civil servants having varied greatly between 2020 (1.1% for magistrates and 1.6% for civil servants), 2021 (1.5% for magistrates and 1.8% for civil servants), 2022 (1.4% for magistrates and 2.1% for civil servants) and 2023 (1.4% for magistrates and 2.0% for civil servants), the number of sick people only increased (without there being any numerical news from the Brazilian Judiciary about magistrates and civil servants who would have been cured), while, in the same period, goal 2 of the Judiciary (to judge the oldest case, which is not set by the CNJ, by magistrates, by civil servants and, much less, by Brazilian citizens, but by the presidencies and the ombudsmen of the twenty-seven Courts of Justice of the Brazilian States, by the five Federal Regional Courts and twenty-four Regional Labor Courts) varied from one segment to another, to the point that, for example, it is up to the state judges of the first instance to judge: at least 80% of the cases distributed up to 12/31/2016 and 90% of the cases distributed up to 12/31/2017 in the Small Claims Courts and Recourse Panels [28]; to judge at least 80% of the cases distributed up to 12/31/2017 and 90% of the cases distributed up to 12/31/2018 in the Small Claims Courts and Recourse Panels [29]; to judge at least 80% of the cases distributed up to 12/31/2018 and 90% of the cases distributed up to 12/31/2019 in the Small Claims Courts and Recourse Panels [30]; and judge at least 80% of the cases distributed up to 12/31/2019 and 90% of the cases distributed up to 12/31/2020 in the Special Courts and Recourse Panels [31].
For labor judges, the following first-instance targets were set: to judge 92% of cases distributed by 12/31/2018 (in 2020); to judge 93% of cases distributed by 12/31/2019 (in 2021); to judge 93% of cases distributed by 12/31/2020 [30]; and to judge 93% of cases distributed by 12/31/2021 (in 2023). The targets were set at a lower level and without any increase from 2020 to 2023 for the State Court, while for the Labor Court, the higher level was set, with an increase from 2020 to 2021, remaining the same in the remaining years.
This seemingly disconnected scenario of mental illness, Judiciary goals, productivity, and AI projects is shown to be interconnected when one realizes that the AI project enables greater productivity; greater productivity makes the manager increase the goals, forgetting that there is still a lot of work that is done entirely by human beings; and the excessive demand for goals, as in cases involving moral harassment, can generate anxiety, stress, and even depression. This interconnection is also identified when art. 2, IV, of Resolution No. 372 of the Superior Council of Labor Justice [32], of November 24, 2023, established, among others, that supplementary compensation for the procedural accumulation of overtime would only be paid to judges if they fully and cumulatively complied with Goals 1 and 2 of the CNJ in the previous year.
These goals are to judge more cases than those distributed and to judge older cases. This regulatory device, despite being aligned with the principle of efficiency and effectiveness of the provision of jurisdiction, links the extraordinary backlog of cases and, consequently, the payment of compensation corresponding to the fulfillment of the goals set by the Judiciary, transforming the danger that not only the Labor Court, but the entire Brazilian Judiciary runs: of mechanizing/industrializing judicial decisions to merely promote intense and numerical productivity in the resolution of cases so that the goals are met and compensation is paid to the judges.
This is because the parallel between the exponential growth of new lawsuits filed with the Brazilian Judiciary, the need for increased productivity without increasing expenses, and the growth in mental illness among judges and civil servants, with symptoms of the most commonly diagnosed illnesses (such as, for example, constant and exaggerated worry, feeling short of breath, anxiety, depression, insomnia, difficulty concentrating, sadness, and disinterest in activities), indicates that judges who are worried, anxious, depressed, and have difficulty concentrating, in addition to not taking time off work to take care of their mental health, will prefer to use the minutes created by AI projects in order to avoid not receiving compensation for the procedural accumulation of overtime.
Furthermore, judges will likely put even more pressure on already ill civil servants to produce more, in order to ensure that targets are met, compensation for the accumulation of overtime paid in the proceedings, and the policy of achieving targets, including for purposes of assessing merit for promotion of judges and access to the Courts of Second Instance, pursuant to Resolution No. 106 of the CNJ, of April 6, 2010 [33]. Despite the fact that the targets were set without considering the local difficulties of each jurisdictional unit, such as, for example, the distance of the Court from the State Capital; the lack of interest of technical experts and doctors in working in distant locations, given the difference between what was needed to be spent and the value of the arbitrated fee; the tiny number of civil servants relocating to units closer to the State Capital; and the lack of interest of judges in settling in locations that are difficult to access.
Obviously, the central problem is not the setting of goals, much less the use of AI. Nor is it being argued that AI projects are causing illness among judges and civil servants. The problem is that, in addition to many managers not seeing the human limits of judges and civil servants (in terms of production, since they are people who have families [children], study and have the right to rest), the Balanced Scorecard (BSC) methodology adopted by the CNJ [34] for the purpose of setting goals is not in line with the provision of judicial protection, with the exponential reality of judges with mental and behavioral disorders, and even less with AI projects based solely or mainly on increasing productivity.
And it is not consistent because, while the Balanced Scorecard (BSC) methodology, based on 4 perspectives (financial, customers, internal processes and innovation and learning) and adopted by the CNJ, aims, among others, to increase business profitability, increase the profit margin on sales, reduce repair costs, build customer loyalty, diversify products/services, develop new distribution channels, improve product quality and create a culture of teamwork, the Brazilian Judiciary, in addition to not aiming to obtain or increase profit or sales and, much less, vary or change products, intends to pacify society, resolving disputes brought in Court, through sentences handed down by a single person in the first instance, reducing conflicts.
It is not appropriate because the Balanced Scorecard (BSC) methodology was created for industries and not for the Judiciary, whose main purpose is to promote Justice. It is not appropriate because the industrialization of decisions in favor of increasing productivity, through the insertion of AI innovations, in a scenario of mentally ill judges and whose compensation for the extraordinary backlog of cases is only paid if the CNJ’s goals are met, will encourage a mechanical work of “copy and paste”, putting at risk the social function of the Brazilian Judiciary as the guardian of the Constitution.
In fact, the risk to democracy is great because the mechanization of decisions, combined with the payment of compensation, as long as the goals are met, makes the leaders of the Judiciary forget that there are people behind the processes. Citizens who want and expect to be judged by sane people. Magistrates who read and know the case deeply (and not superficially), so that they enforce the country’s Constitution, even if it is “against” the interests of the majority of citizens (countermajoritarian principle). Magistrates, therefore, who should not use AI to survive the enormous number of processes, but who should have their greatest asset as their sensitivity and their greatest value as their humanity, as recalled by Guilherme Guimarães Feliciano and José Antônio Ribeiro de Oliveira Silva [35], including to treat civil servants with dignity and humanity.
This is not about sentimentality, daydreaming or freshness. The Health Panel of Magistrates and Employees of the Judiciary for the years 2020 to 2023 shows that the Court of Justice of the State of Pará (TJPA), the Regional Federal Court of the 2nd Region (TRF2), the Superior Labor Court (TST), and the Regional Labor Court of the 21st Region (TRT21) were the bodies that appeared most often in 1st place in the statistics, as being the place with the most magistrates and employees with mental disorders. It also shows that the Court of Justice of the State of Rondônia (TJRO), the Regional Labor Court of the 10th Region (TRT10, Federal District and State of Tocantins) and the Regional Labor Court of the 17th Region (TRT17, State of Espírito Santo) had ranked 1st in the same statistics, respectively, in 2020 (corresponding to 3.3%), in 2021 (corresponding to 4.0%) and in 2023 (corresponding to 3.2%).
It is clear that, despite the CNJ having introduced Resolution No. 207, of October 15, 2015 [36], to establish the Policy of Comprehensive Health Care for Magistrates and Civil Servants of the Judiciary, including guided by the principle of democratization of the governance of this Policy and of health actions and by the principle of intra and intersectorality of health actions (art. 3, IV and V), for health units, among others, to produce and analyze statistical data (taking them as subsidies for proposing new actions in the health area), the CNJ’s 2023 research panel at no time revealed any AI project whose purpose and reason for existing was first and foremost to improve the mental condition of magistrates and civil servants. In addition to the fact that AI projects were not being developed with the aim of improving the mental health of judges and civil servants, the fact that TJPA, TRF2 and TST remained as some of the bodies where the largest number of sick people were concentrated demonstrated that the analysis of statistical data on the illness of judges and civil servants was not producing major improvements in the area of health.
Far from it, the comparison between the fact that the TST remained in first place as the Superior Court with the most sick people in 2020 (corresponding to 2.7%) and in 2023 (corresponding to 3.5%), as well as the circumstance that the TJRO remained as the State Court with the most sick people in 2020, and even so, they were winners, respectively, of the 2020 CNJ Quality Award Diamond, the 2023 CNJ Quality Award Diamond and the 2020 CNJ Quality Award Diamond, suggested that the awards would have been won, yes, by the merit of the institutions, but also at the cost of the illness of magistrates and employees. He also pointed out that the fight for established goals (including obtaining the CNJ Quality Award), according to the Balanced Scorecard (BSC) methodology, as well as the analysis of the numbers raised by the Justice in Numbers Report, was being used to justify the use of AI projects to mechanize decisions and rulings. Especially considering the lack of published data that revealed the existence of any type of containment that, for example, blocked possible drafts of orders, decisions, sentences and rulings that had merely been copied from the AI tool and pasted into Pje.
The Brazilian Labor Court seems to have noticed the imbroglio involving the goals, compensation for the extraordinary backlog of cases, and the illness of judges and civil servants, because despite having incorporated three artificial intelligence tools in early February 2025, including Chat-JT, on February 7, 2025, the President of the Superior Council of Labor Justice (CSJT), Minister Aloysio Corrêa da Veiga, distributed a proposal to amend articles 2 and 3 of Resolution no. 372 of November 23, 2023 [32] (which is being processed under no. 1000060-52.2025.5.90.0000), under the argument, among others, that item IV of art. 2 was producing “distortions and deleterious effects from the point of view of encouraging the dedication of judges and equalizing the workload”. This is because the “criteria” for compliance with Goal 1 made it practically impossible for Courts with a large number of procedural movements to comply with it, or, “at least, excessively difficult and potentially harmful to health”.
However, even though the CSJT recognizes that a different mechanism should be introduced in these overload situations, the proposed solution (still pending decision until the end of this research) of imposing an additional annual procedural volume for the judicial unit with reduced procedural movement (understood as the one that receives less than 70% of the average number of cases distributed to the other Courts of the respective Court) annually, so as to allow for a total productivity equivalent to at least 70% of the average number of cases, would not resolve the embarrassment involving the goals, compensation for the extraordinary backlog of cases and the illness of judges and employees.
This is because the attempted solution would continue to be based solely on increasing productivity, without, however, a proposal that would combine the necessary observance of the Principle of Efficiency and the Principle of the Biopsychosocial Approach to the Health/Illness Process, so that the number of judicial acts would increase in the same proportion as judges and civil servants would be cared for. It is not a matter of caring only for the disease. It is important to see that the scenario outlined will remain the same, perhaps with an increase in mental illness among judges and civil servants, because the search for numbers will continue to disregard, for example, how a single Public Civil Action with 45 requests, in the supposedly low-volume Court, causes disruption and delays in the delivery of the jurisdictional service. A scenario that will maintain the total disregard for the quality of the sentences and rulings handed down. A scenario that will remain favorable for sick judges to choose to copy and paste draft sentences into the electronic process (Pje).
Thus, given this prognosis, even though statistical surveys are indeed important for judicial governance and the inclusion of artificial intelligence in the Judiciary, by itself, does not cause illness among judges and civil servants, as long as the mental illness of judges and civil servants is growing exponentially and compensation for the extraordinary backlog of cases or for any other reason is linked to compliance with the goals set by the Judiciary, the link between the goals of the Judiciary, productivity and AI projects may indeed aggravate the mental disorders observed and increase the number of people affected. This is because male and female judges, as well as male and female civil servants, will be urged to work, even indirectly, even when sick, in addition to the mechanization of sentences and rulings, as well as the complete disconnection from the quality of the delivery of decisions, contributing to the maintenance of the image of indignity, of the uselessness of work and that “the work undertaken has no reason to be” [35].
4. How Can Productivity and the Use of AI Be Reconciled
without Causing Mental Illness among Judges and Civil Servants?
In fact, it is well known that the exercise of jurisdictional activity imposes restrictions and personal demands that are different from those imposed on citizens in general, as established by art. 16 of the Brazilian Judiciary Ethics Statute [37], so that the AI project to contribute to increased productivity, especially in view of the frenetic growth of new cases, would not in itself generate any inadequacy, much less any personal demands that are different from those imposed on citizens in general.
On the contrary, AI meets the normative force of the Principle of Efficiency, insofar as the tool allows the Judiciary to act in such a way as to obtain the best with the lowest possible cost. It allows the Brazilian Judiciary to use the available resources in the best way, with rationality, economy and productivity, contributing to the services provided being of high quality and meeting the needs of the population in a satisfactory manner.
On the other hand, how should the public administration be held accountable for the results of its actions and for the harm caused to the population? Therefore, it does not seem reasonable or even proportional that Brazilian companies should include the assessment of psychosocial risks in the Occupational Health and Safety (OHS) management process, in order to identify and manage risks such as stress, harassment and excessive mental workload, as part of the measures to protect the health of workers, and that the State itself, through its Courts, should increase productivity targets and link the payment of compensation, to which Brazilian judges are entitled, to the fulfillment of targets that the managers themselves know will be difficult to achieve, without any analysis of the impact that this will have on the health of judges and employees.
It is not reasonable, in fact, because in addition to the fact that the payment of compensation is linked to the achievement of goals, which in itself generates stress, anxiety and excessive mental burden, it is clear that in a scenario where mental illness is growing at a frantic rate, the inclusion of AI projects will be for good and for bad at the same time. For better, because it will allow judges, even if they are sick, to draft decisions, sentences and rulings (albeit with the help of civil servants) to meet the goals imposed by the Judiciary, so that they will not need to take time off work. For worse, because, in addition to not taking care of their mental health, they will demand that civil servants, who are also sick, work more (albeit without quality) so that the goals are achieved, creating a vicious circle in which judges and civil servants will become sicker because, respectively, they no longer see any meaning or importance in their work and because they suffer harassment in the name of increased productivity.
An institution whose function is to ensure respect for the rights of Brazilian citizens, through the complex act of judging, imposing punishments for those who commit crimes and resolving conflicts of interest brought to light, cannot allow AI to be used to teach and/or encourage insensitivity and coldness for mentally ill judges and civil servants. This is because the comparison between the criterion used by the factory of “increasing business profitability” through increased productivity shows that AI will be used by many managers as an argument to further increase targets and, consequently, their demands, in the sense that the numbers must be met with the same agility and precision achieved by AI projects, highlighting the vulnerability of judges and civil servants who need more attention [38].
Worse still, even with the number of sentences increasing, there will be a social risk that the resolution of judicial conflicts will end up being compromised, since the solution presupposes sufficient time to read the case file, hear parties and witnesses, reflect and look into the various consequences that may be generated by the sentence or ruling. This situation is not compatible with the haste of those who are more anxious or with the speed inherent in AI projects.
Otherwise, what is meant is that the numbers in the reports of AI projects will not tell the truth. They will not effectively correspond to the work carried out, which is not always of high quality, and even less to the work that generates in society the feeling that it is not worth committing crimes and/or failing to comply with the standards establishing rights and guarantees. After all, none of the panels surveyed analyze the quality of the decisions and/or demonstrate any containment mechanism to prevent the draft eventually created by the AI from being simply pasted onto the PJe platform. None of the panels, especially on AI, showed that the Courts were concerned with increasing the quality of judges’ decisions and/or reducing the number of judges and employees affected by mental illness, in the same proportion and concern as there was about the need to increase productivity.
The issue is not easy to resolve and requires joint work by several professionals, since a simple contingency, for example, prohibiting a mentally ill judge from working in the PJE with AI, could worsen the situation of the judge’s mental and behavioral disorders and/or generate enormous discussion about alleged discrimination resulting from the confirmed illness. In addition, some judges could claim that the inhibitory measures imposed would be preventing judges from accessing the judicially recognized right to compensation, precisely to punish those with mental illness who had not produced to the same extent as other judges.
For this reason, the best solution would be to link the goals of the Judiciary not simply to the number of sentences/rulings issued, but to the quality of the decisions that were issued. This quality could be measured, for example, by the sentences and rulings that were annulled due to lack of grounds and/or by the number of appeals for clarification that were accepted. This quality could also be verified by the judges’ respect for the binding precedents of the STF and other Courts and by the explanation in cases of distinguishing (situations in which the precedents should not be applied due to different factual issues).
However, even if the Balanced Scorecard (BSC) methodology is not abandoned by the CNJ, the inclusion of AI projects could be used in a manner compatible with the need to increase productivity, provided that, instead of the AI delivering a draft sentence or ruling, it were limited to indicating possible inconsistencies between the theses presented by the initial claim and the defense, when in parallel with documents and oral evidence produced, since, in such a case, the judges could be familiar with the case files more quickly and study in depth what was contradictory. This compatibility would also occur if there were a containment generated by the AI preventing any draft originating, for example, from the CHAT-JT (adopted by the Labor Court), from being accepted by the electronic process platform, since such a political choice would contribute to reducing the risks of impact on the social function of the Brazilian Judiciary as the guardian of the Constitution.
Therefore, these compatibility issues would require further studies by the Brazilian Judiciary and a courageous willingness by the Court leaders to recognize that AI would only be inconsistent with productivity if compensation continued to be paid to judges if the goals set by the Judiciary were met, since, in such a situation, the statistical numbers of illness among judges and civil servants that the Judiciary itself compiled year after year would be ignored.
After all, not even Article 16 of the Ethics Statute [37] could be invoked to justify the continued exponential increase in illness among judges and civil servants, since the restrictions and personal demands that are different from those imposed on citizens in general, arising from the exercise of judicial activity, did not authorize or permit the State itself, through its Courts, to make political choices that would make judges and civil servants ill, under the pretext that the public interest should prevail.
5. Concluding Remarks
Just as innovations brought about by artificial intelligence have invariably invaded Brazilian reality, they have also been introduced into the Judiciary, so much so that the CNJ’s Justice 4.0 panel, updated as of February 8, 2025, revealed that 62 Courts had 140 AI projects under development, with 43 of these projects already in use.
Of these 62 Courts, 41 Courts in the country (43.61%) began using AI tools in order to increase productivity (volume/time), despite the fact that the TRT of the States of Pará and Amapá and the TRT of the State of Paraíba were awarded, respectively, the 2024 CNJ Quality Award for Excellence and the 2024 CNJ Quality Award for Diamond, even though they did not have an AI tool registered in the aforementioned panel and had met the requirements, including those related to the productivity axis.
The same CNJ statistical panel showed that the TJSP, despite not having won any CNJ Quality Award in 2024, had judged more cases in 2024 than in 2023, 2022 and 2021, even without having an AI technology project under development and/or use aimed directly at productivity, showing that there were bodies of the Judiciary that had improved their performance through means other than AI, not specified in the reports.
However, the fact that Courts had achieved greater productivity without the use of any AI tool did not mean that the AI innovations introduced in the Judiciary were unnecessary, since the number of lawsuits grew vertiginously year after year, expenses only increased (although not at the same speed at which the lawsuits were distributed) and the collection of revenues by the Brazilian Judiciary could not keep up with the growth in expenses.
In fact, considering that AI would hardly be able to copy all the characteristics of human beings working in the Judiciary, it was necessary to recognize that artificial intelligence should be used as a tool to enhance the functional skills of judges and civil servants and not as a means of disqualifying them or stifling their innovative capabilities.
Proof of this is the fact that the use of AI projects by the Brazilian Judiciary, most of which were created by the Tensorflow Frameworks, using the Bert language model, the Lightgbm, Regex and Xgboost algorithms, as well as the Scikit Learn library, proved that AI had increased the efficiency and agility of procedures, the precision and consistency of repetitive tasks, improved decision-making and reduced errors.
Otherwise, the Health Panel of Magistrates and Employees of the Judiciary Branch for 2020, 2021, 2022 and 2023 showed that the number of magistrates and employees who had to leave public service due to mental and behavioral disorders only increased from one year to the next, at the same time that productivity increased and the goals set by the CNJ were raised and/or kept high.
The scenario of mental illness, goals of the Judiciary, productivity and AI projects, seemingly disconnected, proved to be interconnected when it was realized that the AI project enabled greater productivity; greater productivity made the manager increase the goals, forgetting that there was still a lot of work that was done entirely by human beings; and the excessive demand for goals, as in cases involving moral harassment, could generate anxiety, stress and even depression. This interconnection was also identified when art. 2, IV, of Resolution no. 372/2023 of the Superior Council of Labor Justice, established, among others, that supplementary compensation for the procedural accumulation of overtime would only be paid to judges if they fully and cumulatively complied with Goals 1 and 2 of the CNJ in the previous year.
This situation has shown that magistrates who are worried, anxious, depressed and have difficulty concentrating, in addition to not taking time off work to take care of their mental health, will prefer to use the minutes created by the AI projects, in order to avoid not receiving compensation for the procedural accumulation of overtime work.
The central problem, therefore, is not the setting of goals, but the use of AI. It has also not been proven or even argued that AI projects are causing illness among judges and civil servants. The problem is that, in addition to many managers not seeing the human limits of judges and civil servants (in terms of production, since these are people who have families [children], study and have the right to rest), the Balanced Scorecard (BSC) methodology adopted by the CNJ for the purpose of setting goals is not in line with the provision of legal protection, because while the methodology aims, among others, to increase business profitability, increase the profit margin on sales, reduce repair costs, build customer loyalty, diversify products/services, develop new distribution channels, improve product quality and create a culture of teamwork, the Brazilian Judiciary, in addition to not aiming to obtain or increase profits or sales and, much less, vary or change products, intends to pacify society, resolving disputes brought before the Court, through sentences handed down by a single person in the first instance, reducing conflicts.
The issue is not easy to resolve and requires joint work by several professionals, since a simple contingency, for example, prohibiting a mentally ill judge from working in the PJE with AI, could worsen the judge’s mental condition and/or generate enormous discussion about alleged discrimination resulting from the mental illness confirmed. In addition, some judges could claim that the imposed inhibitory measures would be preventing judges from accessing the judicially recognized right to compensation, precisely to punish those with mental illness who had not produced to the same extent as other judges.
The best solution would be to link the goals of the Judiciary not to the number of published sentences/rulings, but to the quality of the decisions handed down, with the quality being determined, for example, by the sentences and rulings annulled due to lack of grounds and/or by the number of appeals for clarification accepted, as well as by the magistrates’ respect for the binding precedents of the STF and other Courts and by the explanation in cases of distinguishing.
Thus, compatibility with the desired productivity could be achieved if the AI only indicated possible inconsistencies between the arguments presented by the initial claim and the defense, when compared with documents and oral evidence produced, since, in such a situation, the judges could understand the case more quickly, deepening their studies on what was contradictory. Harmony that would also result from any restraint generated by the AI preventing a draft originating, for example, from CHAT-JT (adopted by the Labor Court), from being accepted by the electronic process platform, given that such a political choice would contribute to reducing the risks of repercussions on the social function of the Brazilian Judiciary as guardian of the Constitution.
Compatibilities that would require further studies by the Brazilian Judiciary and courageous willingness by Court leaders to recognize that AI would only not be compatible with productivity if compensation continued to be paid to judges if the goals set by the Judiciary were met, since, in such a situation, political choices would be acting to make judges and civil servants sick, under the pretext that the public interest should prevail.