Artificial intelligence is gradually becoming an ordinary working tool for lawyers. Just a few years ago, the use of AI in court might have seemed a matter of the distant future. Today, lawyers use it to analyze documents and prepare drafts of procedural documents, citizens obtain legal information, and courts use digital tools to analyze judicial practice and process large volumes of data.

Therefore, in my view, the question of whether there is a place for artificial intelligence in justice is gradually losing its relevance. AI has already come to the legal profession. What is far more important now is to determine what exactly can be entrusted to it and where exclusive human responsibility must remain.

For Kazakhstan, this question is no longer theoretical but has quite practical significance. The judicial system uses "Digital Analytics of Judicial Practice," and in 2026 the use of the "Digital Assistant to the Judge" in civil cases was scaled up. Such technologies are capable of analyzing the content of a claim and the parties' arguments, structuring information, formulating the factual outline of a dispute, and finding similar judicial practice. At the same time, the fundamental position of the judicial system is that the digital assistant does not make a judicial decision; the assessment of circumstances and the application of law remain with the judge.

This is the right approach. However, the capabilities of artificial intelligence will expand, and the single general principle—"AI only assists the judge"—may over time prove insufficient. It is one thing when a program corrects a grammatical error or finds a similar court decision in a few seconds. It is quite another when a system analyzes evidence, recommends a particular outcome of a case, or proposes a reasoning for the decision to the judge. Here a simple rule can be formulated: the closer artificial intelligence comes to the exercise of judicial discretion, the higher the requirements for its use must be.

The experience of other countries shows why such rules are necessary already today. In the well-known American case Mata v. Avianca, lawyers submitted to the court a document prepared using ChatGPT. It contained references to non-existent court decisions. The court imposed sanctions in the form of a fine. However, the main lesson of this case is not that lawyers must not use ChatGPT. The problem was different: the lawyers did not verify the information obtained.

Later, similar cases recurred. Moreover, the situation proved more dangerous when AI-generated text contained not an obvious fabrication but a mixture of reliable and unreliable information. Such a document looks professional, contains real cases and correct legal constructions, but among them there may be a fabricated quotation or an incorrect interpretation of a decision. The error becomes significantly harder to notice.

Hence follows perhaps the main rule for the use of AI in the legal profession: the work can be partially delegated to technology, but responsibility cannot.

A lawyer may use AI for initial research, drafting the structure of a document, analyzing a large volume of information, or preparing a draft. But before submitting a document to court, it is the lawyer who must verify the cited legal norms, court decisions, quotations, and factual assertions. The explanation—"artificial intelligence wrote this"—cannot release a lawyer from professional responsibility.

This approach is gradually taking shape in American legal practice as well. The American Bar Association states that the use of generative AI does not eliminate the traditional duties of a lawyer: competence, confidentiality, communication with the client, and responsibility for the legal assistance provided. In Kazakhstan, the Republican Bar Association holds a similar position: AI is regarded as a supplement to a lawyer's professional activity, not as a replacement for his or her legal expertise.

Another practical question arises: should a lawyer inform the court that AI was used? In my view, there is no need to create unnecessary bureaucracy here. If a program corrected grammar, translated text, or helped with formatting, a special notification is hardly necessary. But if AI substantially participated in preparing the legal position, analyzed evidence, or created a significant part of a procedural document, a simple rule could be provided that the person signing such a document confirms that he or she has personally verified the materials prepared with the help of AI and accepts responsibility for their content.

Such disclosure should not be perceived as a sign of the document's unreliability. On the contrary, it can become an ordinary part of professional transparency.

Equally important is the issue of attorney-client privilege. Let us imagine an ordinary situation: a lawyer uploads materials of a criminal case, a client's medical documents, correspondence, or other personal information to a publicly accessible AI service and asks it to prepare an analysis. For the user, this may look like ordinary work with a program, but in fact confidential information is being transferred to an external information system.

Therefore, a simple rule is also required here: protected case materials and information constituting attorney-client privilege must not be transferred to open AI systems unless an appropriate secure mode of processing them is ensured. Kazakhstani legislation already contains general requirements for information protection, and the Law "On Artificial Intelligence" enshrines the principles of responsibility, human controllability, data protection, and confidentiality. Now these general principles must be gradually translated into the language of concrete legal practice.

An even more cautious approach should be taken to the use of AI by the judge himself. There are no serious grounds for prohibiting technologies that help systematize hundreds of pages of a case, compile a chronology of events, find judicial practice, compare documents, or transcribe an audio recording. This can substantially save judicial time.

But there are functions that cannot be imperceptibly delegated to an algorithm: the final assessment of evidence, determination of the credibility of a person's testimony, choice of legal position, and the making of the judicial decision itself. Artificial intelligence does not bear judicial responsibility, does not possess the independence of a judge, and is not capable of replacing human understanding of the specific circumstances of a case.

That is precisely why the European approach treats certain AI systems used to assist judicial authorities in researching and interpreting facts and law and applying the law to specific circumstances as high-risk systems. The Council of Europe also proceeds from the need to preserve human control, transparency, security, and the possibility of verifying the results of AI work.

This raises a difficult question: should a party know that the judge used artificial intelligence in considering its case?

If it is a matter of technical document search or sorting of materials, a separate notification to the parties is hardly necessary. But the situation changes if a system analyzes disputed evidence, predicts the outcome of a case, or proposes a legal position that may influence the decision. In such cases, in my view, a participant in the proceedings should have the opportunity to learn what role the technology played in the consideration of his or her case. Judicial AI must not turn into a "black box" whose output cannot be understood or challenged.

There is another problem that courts will increasingly have to face. Artificial intelligence is capable of creating practically realistic photographs, voice recordings, documents, and videos. Therefore, soon the question will be not only whether a lawyer used AI, but also whether the evidence itself was created or altered with the help of AI.

Here Kazakhstan should act in advance. Where there is reasonable doubt about the authenticity of digital material, the court should have the ability to require the original file, information about its origin, the file's technical data, information about the changes made, and, if necessary, the opinion of a specialist or expert. The emergence of generative AI does not mean that digital evidence can no longer be trusted. It means something else: the origin of evidence becomes as important as its content.

At the same time, it would be a mistake to see artificial intelligence exclusively as a threat. It is capable of substantially expanding citizens' access to justice. A person without legal education can already today ask AI to explain judicial procedure in simple words, help understand a document received, or draft the initial structure of an application. In Kazakhstan, the case of a citizen who used ChatGPT in challenging an administrative fine has already become well known. It would be wrong to say that "ChatGPT won the case"; the decision was made by the court on the basis of law and the circumstances of the case. But the technology helped the person navigate the legal process.

That is precisely why a complete ban on AI in the judicial sphere would be as unreasonable as unconditional trust in its results.

Kazakhstan already has the Law "On Artificial Intelligence" and its own digital judicial tools. The next step, in my view, should be not another voluminous law but a short and clear standard for the use of artificial intelligence by courts, lawyers, and participants in proceedings.

It can be based on several practical rules: the final judicial decision and responsibility for it always remain with the judge; a lawyer is personally responsible for the accuracy of the materials submitted; substantial use of AI must be accompanied by reasonable transparency; attorney-client privilege and other protected data must not be transferred to insecure systems; for disputed digital evidence there must be a procedure for verifying its origin; a participant in proceedings must have the opportunity to learn about such use of AI by the court as could have substantially influenced the consideration of his or her case.

Finally, one more element is necessary-competence. The ability to work correctly with artificial intelligence should gradually become part of the professional training of judges and lawyers. Dangerous is not only the lawyer who blindly trusts AI. A problem may also be a specialist who does not understand at all the technology whose results he or she will have to work with.

The main risk of artificial intelligence for justice therefore lies not in the existence of the new technology itself. It arises when a person ceases to understand its limitations and transfers his or her own professional responsibility to it.

It is precisely this boundary between the capabilities of technology and human responsibility that, in my view, should become the basis for the further use of artificial intelligence in the justice system of Kazakhstan.

Madiyar BURKIT,
Deputy Head of the Department of the Judicial Administration
of the Republic of Kazakhstan for Akmola Region,
Master's student at the University of Southern California (USA),
recipient of the international "Bolashak" scholarship