Debates surrounding the role of artificial intelligence (AI) in judiciary systems have intensified, particularly regarding concerns about whether algorithms might replace judges in drafting legal rulings. Critics highlight the potential risks of a fully automated judgment process devoid of human oversight, yet they argue that this focus often overshadows crucial pre-judgment elements that significantly influence legal outcomes.
According to Estefânia Maria de Queiroz Barboza, a constitutional law professor at the Federal University of Paraná, and Gustavo Buss, a PhD candidate at the same institution, discussions about judicial AI fail to consider vital “accessory” decisions that contribute to the final judgment. These include procedural mandates that can alter how disputes are framed and evidentiary rulings that can be pivotal to the litigants’ cases. They suggest that the backstage activities of algorithmic sorting and precedent selection critically shape the decision-making process long before a judge reaches a verdict.
Barboza and Buss examine Brazil’s efforts to regulate judicial AI amid an increasing caseload, which reached 75.5 million pending cases and 40.9 million new filings in 2025. The National Council of Justice (CNJ) has introduced AI tools, such as Berna, designed to triage and categorize cases more efficiently. In 2025 alone, this tool processed information on 30 million cases and organized 2.5 million of them into 353,000 clusters.
While courts view case sorting as a routine task, Barboza and Buss argue that it inherently involves making judgments about which legal distinctions matter, thus influencing the course of justice. The AI-driven triage process can introduce biases, especially since summaries used for achieving a consensus often lack a comprehensive portrayal of legal reasoning found across different judgments.
Brazil’s regulatory response to judicial AI is notably ambitious. Resolution 615/2025 classifies AI systems based on their function during adjudication, distinguishing between low-risk tasks that involve sorting information and high-risk tasks that result in judicial conclusions. This differentiation mirrors frameworks like the EU AI Act, which categorizes systems assisting judicial authorities based on their functionality.
Barboza and Buss emphasize that decisions made by AI systems prior to human involvement can lead to significant oversight issues. Their concerns are not limited to Brazil; they note similar challenges in the U.S., citing the case of State v. Loomis, where concerns arose regarding proprietary algorithms influencing sentencing decisions.
The authors caution against focusing solely on the final judgments produced by AI systems without considering the interpretive authority exercised in the background. They stress that a transparent and accountable approach to AI in judicial processes is essential and advocate for clear explanations of how AI systems arrive at their conclusions.
As Brazil seeks to regulate the increasing role of AI in its judicial system, the discussion regarding who holds the authority to interpret legal arguments remains pressing. The establishment of a National Committee on Judicial AI aims to oversee the employment of these technologies, though the balance of decision-making power raises questions about accountability among various stakeholders in the judicial process.
