Commentary on Article 16: Ethical Limits and Conditions for a Judge’s Use of Artificial Intelligence

Article 16 establishes the basic ethical framework governing a judge’s use of artificial intelligence technologies. This commentary seeks to elaborate that framework by identifying the conditions of permissible use, the principal risks involved, and the safeguards necessary to preserve the human, independent, and accountable nature of judicial decision-making

This commentary should be read in conjunction with the Code of Judicial Ethics approved by the Decision of the XX Regular Congress of Judges of Ukraine of 18 September 2024 “On Approval of the Code of Judicial Ethics” (https://zakon.rada.gov.ua/rada/show/n0001415-24#Text) and the Commentary to the Code of Judicial Ethics approved by the Council of Judges of Ukraine by Decision No. 14 of 2 March 2026 (https://rsu.gov.ua/uploads/article/komentar2603061320pbkce-af35f4321b.pdf). It is intended as an interpretative and professional guidance text and should be understood in light of those documents.

Article 16

A judge’s use of artificial intelligence technologies is permissible provided that it does not affect the judge’s independence or impartiality, does not concern the assessment of evidence or the decision-making process, and does not violate the requirements of the law.

The modern world is characterized by the unprecedented penetration of digital technologies, in particular artificial intelligence and machine learning technologies, into all spheres of human activity. The judiciary, as one of the key state institutions entrusted with ensuring the protection of human and civil rights, freedoms, and interests, and with upholding the rule of law, cannot remain apart from these global processes. AI technologies offer cutting-edge tools capable of potentially transforming traditional approaches to the administration of justice and enhancing its efficiency, speed, and accessibility.

The main purpose of the article under commentary is to establish ethical standards that would enable judges to make use of the advantages of AI technologies while minimizing potential threats to the fundamental principles of judicial proceedings.

In practical terms, the implementation of this article must ensure that the use of AI technologies does not undermine the fundamental principles of judicial independence and impartiality in the administration of justice; preserves the judge’s exclusive competence with regard to the assessment of evidence and the judicial decision-making process, thereby preventing the delegation of these essential judicial functions to technology; and guarantees that the use of AI in a judge’s professional activity takes place in strict compliance with the requirements of applicable legislation and does not violate the constitutional rights and freedoms of individuals and citizens.

At the same time, the article under commentary performs an important protective function, namely:− preventing the delegation of judicial powers. Judicial powers, such as the assessment of evidence, the interpretation of law, and the rendering of decisions, constitute an integral part of the judge’s constitutional status and personal responsibility. Their delegation, even in part, to AI technologies is impermissible, as this would run counter to the principle that justice is administered exclusively by courts and to the judge’s personal responsibility for every decision rendered by him or her (analogously to the prohibition on delegating the functions of courts to other bodies enshrined in Article 124 of the Constitution of Ukraine);− preventing the loss of the judge’s subjective inner conviction. A judge’s inner conviction, formed on the basis of the direct examination of all the circumstances of the case and the evidence, guided by law, is the foundation of fair adjudication. Excessive reliance on AI, or on its potentially “persuasive” conclusions or suggestions, may blur or displace the judge’s own conviction;− preventing dependency on digital technologies. This refers not only to technical dependency, such as the inability to perform work without a particular tool, but also to the possible psychological influence of the “authority” of technology. If a judge becomes excessively dependent on AI in formulating thoughts, searching for solutions, or even preparing draft documents, this may call into question the judge’s actual independence and impartiality.

Analysis of the conditions for the permissible use of AI technologies by a judge

The article under commentary establishes four mandatory and interrelated conditions, only the simultaneous observance of which permits a judge to use AI technologies ethically. The absence of even one of these conditions renders such use unacceptable.

The first condition: “does not affect the judge’s independence and impartiality.” This is a fundamental requirement that protects the very essence of justice. Judicial independence is an inherent component of judicial status and a key guarantee of a fair trial. It is enshrined in Articles 126 and 129 of the Constitution of Ukraine, elaborated in the Law of Ukraine “On the Judiciary and the Status of Judges,” and emphasized in a number of provisions of the Code, in particular Articles 1, 2, 5, and 6. The concept of judicial independence encompasses both the judge’s individual independence, meaning freedom from any pressure, influence, instructions, threats, or interference in decision-making, and the institutional independence of the judiciary as a whole. In the context of AI, a judge must be free from the influence not only of traditional actors, such as public authorities, parties, or the public, but also from influence that may indirectly emanate from AI tools or their developers.

Preserving inner conviction. Article 6 of the Code emphasizes that a judge administers justice on the basis of his or her inner conviction. The use of AI must not replace that inner conviction, which is formed through a comprehensive, complete, and objective examination of all the circumstances of the case. AI performing an analytical or predictive function may create the effect of a “suggested decision” (or even lead to the use of an output formed on the basis of a model hallucination), where the judge, without realizing it, accepts the algorithm’s suggestion. For example, if a judge uses a tool that predicts “likely case outcomes,” or employs semi-automated templates to formulate the operative part of a decision, this may result in the loss of independent thinking. This effect is often caused by cognitive biases, whereby the first answer suggested by AI subconsciously influences the subsequent decision, and the judge tends to assign greater reliability to technologically generated conclusions than they actually possess.

Avoiding improper dependency. A judge must not become dependent on specific AI tools or their developers. The choice of tools, where such a choice is made, must be objective and directed exclusively toward improving the efficiency of justice rather than serving anyone’s commercial or other interests.

The risks of algorithmic bias in AI constitute one of the most significant threats to impartiality. AI systems learn from data that may contain historical, social, cultural, or other biases. If such biases are not identified and neutralized at the stage of AI development and testing, the system may reproduce or even amplify them in its conclusions or recommendations. This may lead to discriminatory consequences, which directly contradict Article 9 of the Code, obliging a judge to refrain from discrimination on any grounds. When using AI, a judge must be aware of this risk and take steps to minimize it, for example by verifying the provenance of the AI’s training data, where possible, or by adopting a particularly critical attitude toward its outputs. If an algorithm is trained on biased data, for example with regard to categories of claimants or types of cases, even an auxiliary AI function, such as the selection of relevant case-law, may influence the judge.

The effect of automation bias lies in the fact that a judge may place excessive reliance on AI-generated conclusions, perceiving them as more objective than his or her own judgments. Such trust in AI creates a psychological risk of unconsciously accepting biased recommendations. One specific manifestation of this is the influence of the “authority” of technology: a judge may be impressed by the supposed particular reliability or scientific character of AI conclusions solely because of their technological origin. This may reduce the level of critical analysis and contradict the judge’s duty to render decisions exclusively on the basis of his or her own assessment of the evidence and the law. A judge must be aware of these risks and actively counteract them.

Transparency and explainability are key prerequisites for the impartiality of a judicial decision. If AI functions as a “black box,” that is, without any possibility of understanding the internal logic by which it arrives at a conclusion, there is a risk that the judge will be unable to explain even to himself or herself the grounds for those conclusions. This is why algorithmic transparency is so important. The impact of opaque algorithms lies in the fact that the lack of understanding of the principles by which AI generates recommendations deprives the judge of a real opportunity to assess them critically, modify them, or reject them. This calls into question the independence of the judicial decision, precludes external scrutiny, and reduces confidence in the outcome of the proceedings.

The second condition: “does not concern the assessment of evidence.” The assessment of evidence is a central element of the judicial function, involving the determination of its relevance, admissibility, reliability, sufficiency, and interrelationship for the purpose of rendering a reasoned decision. This activity cannot be delegated to AI.

The essence of assessing evidence is not merely a technical operation but a complex cognitive process, involving analysis of the content of the evidence, its source, the circumstances in which it was obtained, comparison with other evidence, and consideration of the parties’ positions. It also entails the application of legal consciousness and life experience.

The limitations of AI in assessing evidence. AI may efficiently process large volumes of information, identify trends, and detect connections, but it is incapable of understanding context as a human being does; it lacks the capacity for empathy; it cannot assess the sincerity or insincerity of a witness; and it cannot take into account cultural or social nuances that may be significant for understanding evidence.

Permissible technical assistance by AI. AI may be used as an auxiliary tool for:− organizing and systematizing evidence, for example, creating a chronology of events on the basis of documents, indexing large arrays of textual evidence to facilitate search, or identifying duplicates;− analyzing structured data, for example, analyzing financial transactions for atypical operations in cases concerning economic crimes;− identifying certain objects in photographic or video materials, for example, facial recognition or recognition of licence plate numbers, while the conclusion as to what the presence or absence of that object signifies must be made by the judge.

The impermissibility of automated determination of the reliability or importance of evidence. Any attempt to use AI to determine directly whether evidence is reliable or how important it is in comparison with other evidence constitutes a violation.

The third condition: “does not concern the decision-making process.” The rendering of a judicial decision is an act of applying the law to established facts that concludes the determination of the case on the merits and contains an authoritative pronouncement concerning the parties’ rights and obligations or a person’s guilt. This process is exclusively human and judicial in nature.

Responsibility for the judicial decision. Only the judge bears full personal responsibility for every decision rendered by him or her. Delegating decision-making to AI would amount to renouncing the judicial function and responsibility, which is incompatible with judicial status and the judicial oath (Article 1 of the Code).

The intellectual and volitional nature of decision-making. The process of rendering a decision requires not only knowledge of the law, but also its interpretation, an understanding of both the letter and the spirit of the legal norm, the principles of fairness and proportionality, consideration of all the circumstances of the case, and the individual characteristics of the parties. AI, even in its most advanced form, is incapable of such a comprehensive intellectual and volitional activity.

The secrecy of judicial deliberations. It is important to bear in mind that procedural law prohibits a judge, in the course of rendering a judicial decision, from discussing the circumstances and materials of the case, the application of legal norms and judicial practice in that case, as well as the content of the judicial decision, with any persons other than the composition of the court hearing the case; judges have no right to disclose the course of deliberation and decision-making; and deliberation and decision-making take place on court premises. Even where technology is used to prepare a general draft decision, this must not replace the discussion of each judge’s position in cases of collegiate adjudication. In the case of adjudication by a single judge, the use of AI must not create the effect of an “external interlocutor” influencing the judge’s inner conviction.

The fourth condition: “does not violate the requirements of the law.” This condition is general in nature and encompasses all aspects of legal regulation.

Constitutional guarantees. The use of AI must not only refrain from violating, but must also ensure the realization of, the constitutional rights and freedoms of individuals and citizens. Particular attention should be paid to the right to a fair trial, the right to personal data protection and respect for private and family life, as well as to guarantees of equality and non-discrimination.

Procedural legislation. The use of AI must be strictly integrated into, and comply with, the requirements of the Commercial Procedure Code of Ukraine, the Code of Administrative Procedure of Ukraine, the Criminal Procedure Code of Ukraine, the Civil Procedure Code of Ukraine, the Code of Ukraine on Bankruptcy Procedures, and the Code of Ukraine on Administrative Offences. AI may not substitute the principles of dispositiveness and adversarial proceedings established by these acts, nor the guarantees of party participation and appeal. In addition, a judge’s use of AI must be consistent with the provisions of the Law of Ukraine “On the Judiciary and the Status of Judges.”

Information legislation. It is important to take into account the requirements of the Laws of Ukraine “On Information,” “On Access to Public Information,” “On Personal Data Protection,” “On State Secrets,” and legislation in the field of cybersecurity. This will help prevent unauthorized access to data and its unlawful use.

Legislation in the field of intellectual property protection. It is necessary to ensure that the use of AI complies with legal requirements in relation to the processing of data that may be objects of copyright, and to avoid plagiarism when generating the texts of judicial decisions.

Court records management instructions and other subordinate legislation regulating the procedure for the creation, signing, registration, and storage of draft judicial decisions and other official information.

Ethical boundaries of AI use

The ethical use of AI by a judge is possible only subject to compliance with formal and substantive limitations, directly enshrined in the article under commentary and elaborated in the legislation of Ukraine.

Formal boundaries:− the use only of AI technologies that do not perform legal assessment or interpretation;− the absence of AI involvement in the preparation of the reasoning part of decisions without judicial control;− the prevention of the covert deployment of AI tools in the working environment without proper notification of the judge;− the use only of such AI systems as have undergone appropriate certification or have been recommended by competent judicial authorities, where such procedures have been introduced, or, in the absence of such procedures, such systems as are generally recognized and possess a sufficient degree of transparency and verified reliability for use in the legal sphere, and which also comply with the requirements of applicable legislation on data protection and cybersecurity;− proper documentation of the fact that an AI tool was used and of its specific role in the preparation of materials or the analysis of information, for example in internal notes or, where necessary, in the case file if this has procedural significance. This is necessary to ensure transparency in the decision-making process and the possibility of subsequent control or review.

Substantive boundaries:− AI may not propose the outcome of a case, for example, a decision to grant or dismiss a claim. This is because determining the outcome of a case is the quintessence of judicial discretion, grounded in the judge’s inner conviction formed as a result of a comprehensive analysis of the circumstances of the case and the legal norms, and it cannot be automated or delegated;− AI may not perform the legal qualification of facts. Legal qualification requires a profound understanding of legal norms, their interpretation in the specific context of the case, and the taking into account of legal positions and doctrine, which currently lies beyond the capabilities even of the most advanced AI systems. The judge bears personal responsibility for the correctness of the legal qualification;− a judge may not use AI to determine the priority or reliability of evidence. The assessment of evidence, including its relevance, admissibility, reliability, and interrelationship, is the exclusive prerogative of the judge and is carried out on the basis of his or her inner conviction and direct examination of all case materials. Transferring this function to AI would run counter to the fundamental principles of judicial proceedings.

Relationship with other provisions of the Code

The article under commentary does not exist in a vacuum, but is closely interwoven with other ethical standards established by this Code. Understanding these connections is key to its proper application.

Article 1 (Strict adherence to the principle of the rule of law and the requirements of the law, and to the judicial oath). Any use of AI that is contrary to the law, to the principles of the rule of law, for example by resulting in arbitrary or disproportionate decisions, or that violates the judicial oath, for example through a negligent attitude to verifying AI outputs, is impermissible. Conversely, ethically sound use of AI may contribute to the realization of these fundamental principles

Article 2 (Avoidance of conflicts of interest, resilience to pressure, and independence in decision-making). Independence from AI as an “advisor” is as important as independence from parties or other influences

Article 5 (Judicial independence). The use of AI must not affect judicial independence and, on the contrary, should contribute to ensuring the rule of law in all its manifestations

Article 6 (Inner conviction). AI cannot replace or suggest the logic of judicial reasoning, which is based on a perception of facts unique to the particular judge

Article 7 (Conscientious and competent performance of duties, professional development, and improvement of practical skills). This provision directly indicates the need for a judge to develop continuously. In contemporary conditions, this includes the development of digital literacy, an understanding of the fundamentals of how AI functions, its capabilities, and, equally importantly, its limitations and risks. Conscientious performance of duties presupposes that, where a judge uses AI, he or she does so competently, which requires continuous self-education, critical thinking regarding the capabilities and limitations of technology, and careful verification of AI outputs. The judge bears personal responsibility and may not shift it onto technology

Article 9 (Respect for human dignity and prevention of discrimination). As already noted, the risk of algorithmic bias in AI is extremely serious. The use of AI that results in discrimination on any ground, including race, sex, religion, or property status, constitutes a grave violation of this ethical standard

Article 21 (A judge’s participation in social media and internet forums / Conduct in the digital environment). Although this article regulates a judge’s conduct on the Internet in general, its principles, namely restraint, moderation, caution, and avoidance of harm to authority, are equally applicable to discussion of issues relating to the use of AI where such discussion takes place in the public online space. The principles of safety and caution in interaction with information platforms fully correlate with the ethical requirements governing the use of AI

Analysis of potential risks and limitations:− the risk of the “dehumanization” of justice. Justice is not merely the application of legal norms, but also interaction with people, dialogue, and regard for their emotions, dignity, and individual circumstances. Unlike a human being, AI is incapable of empathy, moral intuition, and understanding the complex social context of human conflicts. Justice requires not only the logical application of the law, but also the capacity for compassion, consideration of individual circumstances, and decision-making that is not only formally correct but also fair in human terms. Excessive reliance on AI may lead to the devaluation of these important aspects of justice;− the problem of accountability and responsibility. If AI makes an error that affects the course of a case, the issue of responsibility, whether of the developer, the court, or the individual judge, is complex and requires clear legal regulation. The article under commentary emphasizes that responsibility remains with the judge;− psychological pressure on the judge. Awareness that his or her decisions will be compared with “optimal” decisions proposed by AI may create additional pressure on the judge and affect his or her independence;− the cost and accessibility of AI technologies. The high cost of developing, deploying, and maintaining high-quality AI technologies may lead to “digital inequality” between courts, which will have consequences for equal access to justice;− the need for constant updating of AI. Legislation and judicial practice are constantly evolving. AI systems require ongoing improvement, which is a complex and resource-intensive process;− ethical dilemmas of programming. In the development of AI, especially AI intended to assist in decision-making, even at preliminary stages, difficult ethical questions arise: which values and priorities should be embedded in algorithms? Who bears responsibility for these “programmed” values?;− the impact on the development of law. If judges begin to rely excessively on AI that is based on existing practice, this may slow down the development of law and the formation of new legal approaches needed to address new societal challenges;− the influence of biased data. AI algorithms trained on data containing hidden biases, for example with regard to certain categories of persons or types of cases, may unconsciously steer a judge toward biased conclusions, even where the judge considers himself or herself completely independent.

Criteria for a judge’s ethical self-assessment when using AI. Before using any digital tool, a judge is encouraged to conduct a brief ethical self-assessment and ask himself or herself, for example, the following questions:− am I using AI exclusively as an auxiliary means rather than as a source of judgment?;− does the AI provide information about the sources it used?;− do I retain control over the AI’s output, including the ability to modify or reject its response?;− does the AI’s response contain bias?

Affirmative answers to all of these questions indicate the ethical permissibility of using AI. If even one of them gives rise to doubt, it is recommended to choose another, more transparent or more controllable tool, or to refrain from using it altogether.