Can a judge use an AI tool to decide your case?

Suppose the judge presiding over your dispute consults an algorithm beforehand that shows him how fellow judges have ruled in similar cases. Is that allowed? The short answer: such a tool does not yet exist for judges in Belgium, but the legal framework is already paving the way for its introduction. And as soon as that tool is available, it will come into direct conflict with several fundamental principles of civil procedure: the independence and impartiality of the judge, the right to be heard, and the obligation to provide a reasoned decision. The key question is not whether the technology is usable, but how it can be deployed without sacrificing the guarantees of a fair trial.

Where does Belgium stand today?

The Central Registry of Judicial Decisions, established by the Act of October 16, 2022. That registry makes judicial decisions available, but does not permit their reuse by computer programs for the benefit of parties to the proceedings or their attorneys. Mass downloading of the data is even expressly prohibited. Only members of the judiciary may, under certain conditions, obtain algorithmic processing of that data to support them in their legal duties (Art. 782, § 4 Jud.C.).

The Constitutional Court addressed this asymmetry in its ruling of January 30, 2025 (No. 9/2025) was essentially accepted. The unequal access between judges, on the one hand, and citizens and attorneys, on the other, was justified by weighing the relevant fundamental rights—including the protection of privacy and personal data—and by the role of each actor in the administration of justice. The Belgian legislature thus opens the door to the regulated reuse of judicial decisions by algorithms, provided that this serves to support members of the judiciary in their statutory duties.

The technology is not yet part of daily practice, but the direction is clear. A tool that provides judges with an overview of case law trends, linked to the relevant statutory provisions, can strengthen legal certainty and reduce the risk of an unintended gap in their reasoning. The guiding principle here is always assistance, not replacement: the human judge remains the decision-maker.

Why an Algorithmic Suggestion Threatens Independence

The first point of contention arises even before the tool is used, namely whether the judge can be compelled to use it. The principle of the separation of powers precludes any general obligation to follow an algorithmic recommendation. A legislature that were to impose on a judge a tool used to define a legal concept or interpret a factual situation would be encroaching on the very core of the judicial function. Mandatory use by operation of law is therefore ruled out; ideally, its use is entirely optional, both in terms of consulting the tool and taking its results into account.

The real danger, however, lies in an actual obligation. Once the tool is available, workload and pressure to increase efficiency may prompt judges to use it to process cases more quickly. Added to this is the general obligation to know and apply the law (iura novit curia): As more powerful tools become available, a mistake becomes less excusable for a “normally prudent and careful” judge. There is a possibility that consulting a decision-support tool will gradually become an implicit obligation.

Once the tool is put into use, a second set of risks arises that directly affects decision-making.

The automation bias and the herd effect

Automation bias is the unconscious tendency of humans to automatically or excessively rely on information provided by a machine. A judge acting in good faith is thus swayed by the persuasive power of the algorithmic suggestion, an effect that is further amplified by confirmation bias: the tendency to trust results that confirm one’s own preconceptions.

In addition, there is the performative or self-fulfilling effect: a “herd effect” that drives users toward conformity. The judge who follows the statistically most likely outcome chooses a reassuring and routine solution that also protects him from criticism in the event of an error. Future case law thus reinforces the majority solution, while dissenting decisions are marginalized. Yet it is precisely these deviations that are essential to the evolution of the law.

Impartiality: Bias and Lack of Neutrality

Technology is never neutral. An AI system depends on the data it is fed and is subject to the societal biases that arise from its training on past case law. Instead of correcting those biases, it can reinforce them. A well-known American example is the COMPAS software, which assigned a higher recidivism risk to people of color, thereby perpetuating a historically entrenched bias.

Furthermore, a judge who receives a suggestion regarding the outcome before the proceedings begin will inevitably form an opinion, even though he is expected to approach the trial without bias. Because the algorithmic result gives the appearance of scientific certainty, this premature judgment is particularly insidious.

Dissent as the Only True Remedy

The traditional remedies of the judicial system—recusal, abstention, and appeals—are not sufficient. Replacing a judge with a colleague who has the same tools at his or her disposal does not solve anything, and the widespread use of single-judge panels further complicates the matter.

The solution lies elsewhere. By its very nature, the algorithm’s suggestion is a factual finding: not applicable positive law, but a statistical synthesis of previous rulings. A fundamental principle of Belgian judicial law, enshrined in Article 1138, 2° of the Judicial Code, prohibits a judge from basing his decision on facts that the parties have not raised in the proceedings. When a judge unilaterally incorporates an algorithmic result into his reasoning—without the parties’ knowledge—he violates the principle of judicial discretion.

Nor can it be remedied by characterizing it as a matter of common knowledge or a fact derived from the case file. An algorithmic prediction is too uncertain and too opaque for that, and the underlying statistics are not accessible to the parties. Consequently, this element must be subject to the parties’ cross-examination.

The AI Regulation supports that approach. AI systems designed to assist judicial authorities in investigating and interpreting facts and the law are classified as high-risk systems (Art. 6, para. 2, in conjunction with Annex III, point 8). Such systems require effective human oversight (Art. 14). In a context where the judge consults the tool only after the proceedings, the only way to ensure that oversight is in place is to include it in the reasoning.

Specifically, what does this mean?

For attorneys and litigants. As soon as such tools are introduced, safeguarding the right to be heard becomes a key focus. This applies equally to the use of generic generative AI by judges, which carries its own risks for the verifiability of the decision and the right to a fair trial—an issue we discussed earlier in Can a judge use AI to reach a verdict, and will you still have a fair trial?. Anyone who suspects that the judge is taking an algorithmic suggestion into account in his or her ruling would be well advised to request in a timely manner that this information—and the weight given to it—be included in the reasoning, so that it can be reviewed by the parties and, where applicable, by the Court of Cassation. Equally important is the possibility of defining the scope of the debate through a procedural agreement. The parties may exclude certain facts or legal remedies from the debate and thereby bind the judge—an application of the principle of party autonomy that also applies to whether or not to follow an algorithmic suggestion. The Court of Cassation recently reaffirmed the principle of the parties’ procedural autonomy (Supreme Court, March 7, 2025).

For magistrates. The secrecy of deliberations protects the judge from external influences, but the influence of an algorithmic suggestion is internal: it operates through cognitive biases. Therefore, the use of the tool and the weight assigned to the result belong in the reasoning, not under the secrecy of deliberation. This does not require a complete presentation of all statistics—a reference to the use of the system, the weight of the result, and a clear explanation of the weighting are sufficient. The prohibition on rendering a judgment by means of a general and rule-based decision (Art. 6 of the Judicial Code) and the prohibition on providing reasoning by reference remain fully in force: a mere reference to an algorithmic result can never support a decision. The judge must form his or her own, independent conviction.

For lawmakers and policymakers. The most balanced solution lies not after but during the proceedings. A truly preparatory process, in which the algorithmic suggestion is shared with the parties in a timely manner, reconciles technological innovation with the right to be heard, procedural efficiency, and the neutralization of cognitive biases through dialogue. The suggestion then becomes one element among many others, and the proceedings are enhanced by the discussion.

Frequently asked questions (FAQ)

Can a Belgian judge consult an AI tool when adjudicating a case today?
That depends on the type of AI. A tool that provides judges with a statistical analysis of case law to support their decisions is not currently operational in Belgium. The legal framework does, in principle, allow for the regulated reuse of judicial decisions by algorithms, but exclusively for the benefit of members of the judiciary and to support their legal duties. Generic generative AI, such as ChatGPT, Claude, or Gemini, on the other hand, is already freely available today and is not regulated by that framework. Its use poses a distinct and acute risk: the fabrication of nonexistent case law, legislation, or legal doctrine.

Is the judge required to state that he used an algorithm?
Yes, if such a tool contributed to the decision. The obligation to provide reasoning and the right of review held by the parties and the Court of Cassation require that the use and weight of the algorithmic result be reflected in the reasoning. The secrecy of deliberations does not preclude this, because it concerns an internal influence on the decision-making process.

What does the AI Regulation say about AI in the judicial system?
The AI Regulation classifies AI systems that assist judicial authorities in investigating and interpreting facts and the law as high-risk systems. This classification is subject to strict obligations, including effective human oversight of the system’s output.

Conclusion

The digital transformation of the justice system affects the structure of the judicial process, the role of the judge, and the judiciary’s role in shaping the law. The question is not whether algorithmic tools will be introduced, but how they can be used without undermining the legitimacy of the process. The answer lies in adversarial proceedings: by introducing algorithmic suggestions into the debate at an early stage, the judge retains his independence and impartiality, the parties can challenge or refine the result, and human, legal reasoning remains intact. Adversarial proceedings are at the heart of the process, and it is precisely this value that provides the leverage to make the change acceptable.