Leotrim Gashi is a judge at the Basic Court of Mitrovica, in the General Criminal Department. He joined the judiciary in 2022, just as the rise of artificial intelligence (AI) had begun. By the end of that year, OpenAI launched ChatGPT — a platform capable of answering almost any question within seconds — gaining around 1 million users in its very first week.

Today, ChatGPT and other AI platforms have become part of everyday life for most people, regardless of profession. However, their use in certain fields raises greater ethical dilemmas than in others. Gashi has often encountered discussions about the role of AI in his daily work.

He explains that the topic raises many questions, which is why he uses AI platforms only as a technical tool, mainly for analyzing legal materials, researching domestic and international court cases, or standardizing decisions of a similar nature.

Similarly, Grok and ChatGPT are the two AI platforms most frequently used by attorney Gerta Ragipi. She finds Grok more convenient, particularly for tasks such as summarizing documents, drafting introductions to case circumstances and other technical aspects.

When handling various civil defamation cases, Ragipi relies on AI platforms to search for similar precedents from the European Court of Human Rights (ECHR).

For data on judicial practice regarding sentencing in criminal cases, Iliriana Çollaku-Tafa, a prosecutor at the Special Prosecution Office in Kosovo, also turned to ChatGPT. However, she was unable to accomplish much, as the platform did not provide reliable sources.

Based on conversations K2.0 has conducted with justice professionals in Kosovo, judges and lawyers often turn to ChatGPT or other AI platforms because there is no accessible platform in Kosovo that provides summaries of local court decisions, international case law, or advanced legal research.

The nature of their work — requiring research into local and international cases similar to those that they handle — makes this process one of the main pillars of their profession. Such research usually takes considerable time, while AI platforms like ChatGPT can complete it in seconds.

Conversations with legal professionals have highlighted both the lack of full digitalization of data and the limited capacity of the judicial system to integrate AI. Such technology could allow courts and judicial administration to automate procedural processes, integrate and harmonize existing data, and establish a unified information base for more efficient consultation, analysis and decision-making.

Only recently has the justice system begun operating several fragmented platforms, which mainly provide basic access to documents and data but still fail to ensure full data integration or facilitate cooperation among judicial actors. One privately developed platform, Lexdoks, offers a database of judicial system data in Kosovo, but its use remains limited due to subscription-based access.

Integrating AI into data access is just one way it could improve efficiency in a judicial system burdened by heavy caseloads, bureaucracy, and outdated procedures — such as the manual transcription of hearings, a task that AI platforms could easily perform.

However, even legal professionals remain cautious about using AI, given both the limitations of these platforms and the sensitive nature of their work.

“Empathy that does not replace humans”

ChatGPT and other AI platforms provide guidance easily by analyzing billions of sentences to predict the next words and generate natural-sounding responses. However, since they do not understand facts in the way humans do, they can be unreliable as guides in legal cases.

Judge Gashi notes that if judges follow AI’s guidance without verifying the accuracy and applicability of such information any further, they risk reaching unfair decisions.

He also expresses reservations about using AI because of the human dimension and sensitivity he believes must remain present in decisions that carry legal consequences.

“Judicial decisions have direct consequences on the lives, freedoms and rights of citizens. Therefore, legal accuracy, professionalism and sensitivity throughout judicial processes are essential in decision-making — values that an AI system, especially in its current form, cannot guarantee,” he said, stressing that the responsibility for any decision ultimately rests with the judge, not AI.

Gzim Shala, a researcher at the Kosovo Justice Institute (KJI), an organization that monitors the justice system in Kosovo, noted a growing interest from within the judiciary in the use of AI. He emphasized, however, that it is crucial to avoid creating the perception that judges’ decisions could be influenced by AI.

While representatives of the justice system in Kosovo currently express reservations about AI, lawyers and judges in the United States, England and other countries are making extensive use of it in decision-making and in drafting pleadings. This trend has raised concerns among international legal systems and professionals about the implications of AI’s interaction with justice.

The use of AI in the justice system is being followed with uncertainty worldwide. In 2024, the European Union (EU) adopted the Artificial Intelligence Act, which classifies the use of AI by institutions in legal disputes with potential consequences as “high risk.” While the document, drafted and approved by EU institutions, does not exclude the use of AI for various administrative tasks, it clearly states that AI should never determine the final outcome. “Decision-making should remain a human-driven activity,” the act emphasizes.

One of the main reasons for discouraging such use is the phenomenon of so-called “hallucinations”, known as AI-generated fabrications. These occur because generative AI (GenAI) models are built on large datasets and predict answers based on patterns, rather than directly verifying facts. When data is limited or a question is unclear, GenAI may fill in the gaps with fictional information.

This is precisely what led the law firm Butler Snow, founded in Mississippi, USA, to issue an apology to the Court after relying on fabricated quotes generated by AI in two separate cases. The firm, along with other law offices that have used AI to draft complaints and decisions, has also been warned of possible sanctions by Judge Anna Manasco, who handled the case.

In the UK, two judges have likewise issued warnings to lawyers against using AI to construct legal arguments or to invent statements from non-existent witnesses, which can result in the submission of false information before the Court.

Meanwhile, Damien Charlotin, a legal researcher in Paris, France, who maintains a database on AI use, identified 95 cases of AI-generated fabrications in the U.S. as of June 2023, and 53 cases in 2025 alone. These statistics reflect instances in which lawyers admitted to using AI to draft legal documents.

Ragipi, a lawyer, added that in her research she has observed that AI often presents the content of legal articles in a way that suits the user, rather than reflecting articles accurately. In other words, instead of showing exactly what a legal provision or source states, AI can generate a version with convincing elements that may not fully comply with the law or the context in which the article is cited. For this reason, Ragipi emphasizes that careful review is essential to verify the accuracy of the information.

A Malaysian court has used AI for the first time to issue a decision, convicting two individuals of drug possession. Specifically, the judges based their sentencing on AI recommendations after providing the system with information on the quantity of drugs seized, the defendants’ profiles, their criminal histories, and socioeconomic data as mitigating or aggravating factors. This move has sparked widespread criticism, with many arguing that AI-based sentencing is unconstitutional, as it is not authorized under the Criminal Procedure Code.

Meanwhile, the increasing use of AI in the US has prompted various court decisions across different states. In 2023, a Michigan ethics opinion advised that judges should distinguish between using AI to make decisions and using it to inform themselves about case circumstances.

While the global use of AI is already accompanied by significant uncertainty and concerns about fairness for all parties, its adoption in Kosovo is even harder to envision, given current assessments of the judicial system’s limited digital capacities.

Fragmented digital platforms

Since 2013, the Kosovo judicial system, with support from international donors, began implementing a project for a central database of judicial data. This system, called the Case Information Management System (CIMS), aimed to modernize case registration and enable automatic distribution of cases to judges and prosecutors.

However,  until 2020, the project faced significant delays. These setbacks were mainly due to limited institutional capacity, poor coordination for integrating electronic data and the lack of sustainable funding for the database. Since 2020, CIMS has been used more extensively, facilitating electronic case registration and automatic distribution to judges and prosecutors.

Through the interconnection of CIMS with the e-Kosova platform, a digital mechanism has also been created to track the status of individual cases in court, allowing online access to case information. However, access to this system is restricted: users must be involved in at least one legal case in CIMS to use it; otherwise, the mechanism remains inaccessible.

Before 2020, case management in Kosovo was largely manual and fragmented. Cases were recorded only in physical books and files — a practice that continues, although future plans aim for full digital recording. Case distribution to judges was handled by court secretariats, and there was no unified standard for registration or monitoring. Citizens had no direct access to information, and central statistics were lacking.

With the introduction of automatic case distribution to the Basic Courts, followed by the Court of Appeal in December 2021 and the Commercial Court in April 2024, this integration made case information electronically accessible. This development laid the foundation for the Open Data platform by the Kosovo Judicial Council (KJC), which became operational in 2022.

The Open Data platform is not an independent system; it operates using data generated by CIMS. The platform compiles statistical reports based on this information and can produce two main types of reports: predefined and dynamic.

Predefined reports provide basic data on the number of cases registered and resolved, the average duration of case processing, the number of sessions held, and the workload of judges. Consequently, if the data from CIMS is incomplete or unstructured, even the predefined reports in Open Data remain limited in terms of the information they provide.

Dynamic reports, on the other hand, allow users to filter and combine data by year, court, department or case type, and to export the results to Excel or PDF formats, enabling more detailed analysis.

However, producing such detailed reports is made difficult because a large portion of judicial documents undergo a manual anonymization process during integration into CIMS to protect the privacy of the parties. Anonymization involves physically masking the identifying elements in the documents, which are then scanned and published as images on the courts’ websites.

Computers cannot read this format, making automatic processing, data categorization and advanced analysis impossible. Even if a decision contains several readable pages, manual anonymizations on just the first and last pages can disrupt automatic processing, including AI-based data generation. This means that the quality and completeness of the information published by Open Data depend directly on the quality of the data integrated into CIMS.