Our view on the use of Artificial Intelligence (AI) in the information environment of the institution

Modernity, functionality, security and long-term sustainability of the technological platform of the IT infrastructure of the Chancellery of the Supreme Administrative Court of the Slovak Republic. Since the establishment of the Supreme Administrative Court, this vision has been gradually reflected in the building of the internal technological environment, while in a partial aspect it can also be perceived by the professional or general public, whether when visiting the court in person or when browsing its website. From this point of view, it is particularly worth mentioning the current search system for the case law of the Supreme Administrative Court, , the ultimate ambition of which is to lay the technological foundation for a new platform, also reflecting the current development of AI (it has been conceived from the beginning with its future use in mind).  

Regarding the general issue of the use of AI in justice, it can first be noted that AI quite naturally penetrates the judicial agenda (indirectly) through submissions of the parties to the proceedings or their legal representatives, for example in the form of proposed means of evidence or interpretation of legal concepts generated by AI (decision of the Municipal Court in Prague, Case no. 10 A 99/2023 of November 9, 2023; judgment of the Regional Court in Trenčín, Case no. 19CoPr/2/2024 of November 21, 2024 and others). Special attention is required in situations where the argumentation presented in the submissions appears convincing and professional at first glance, but may contain references to non-existent case law or other inaccuracies, the verification of which subsequently places a disproportionate burden on the court (resolution of the Constitutional Court of the Czech Republic, Case no. I. ÚS 3004/25 of December 1, 2025; judgment of the Supreme Administrative Court of the Czech Republic, Case no. 3 As 34/2025-80 of April 15, 2025, point 43 and others). 

On the other hand, AI can also represent a practical tool for the judiciary, especially in managing the excessive number of submissions – including those generated by AI – identifying ambiguities and filtering relevant information in a way that avoids influencing the judge in their decision-making. The purpose of the following lines is then to focus precisely on the latter mentioned method of the potential use of AI in the judiciary / in the Supreme Administrative Court (i.e. by the judges or court officials themselves in the exercise of judicial power). 

At the outset, we consider it important to emphasize the fact that the responsibility for the use of AI in the judiciary does not “only” lie with the AI ​​users themselves, but must be perceived at multiple levels. Firstly, it is the state that is responsible for any (possibly incurred) damage resulting from the use of AI in the exercise of judicial power. At the organizational level of a specific institution, it is the statutory body that bears the burden of responsibility, and for certain types of AI systems, it may even have multiple roles within the meaning of Article 3 of the AI ​​Act (and this classification subsequently results in a different range of obligations). And finally, the responsibility also lies with the aforementioned specific user of the AI. 

How does the Supreme Administrative Court and its Chancellery approach the deployment of AI (in the judicial environment)? 

In the first half of 2025, an AI governance team gradually began to be built, addressing the legislative, technical, and security requirements placed particularly on the use of AI in the judiciary. Led by the Head of the Chancellery of the Supreme Administrative Court, this predominantly internal team is composed of individuals covering issues related to the use of AI from the perspective of information technology, adopted European and proposed national legislation, ethical and legal aspects of AI, personal data protection, intellectual property law, AI risks and their management system, and cybersecurity. 

A legitimate question that is being asked by both – technology enthusiasts in the judicial environment and skeptics – is the potential impact of the use of AI on the performance of decision-making activities of courts and its impact on the rights and legal status of individuals and legal entities. Without the need to further elaborate on the question of the impact of AI on the functioning of the legal system as such, which is the subject of deeper analysis in multiple academic texts, the opinion of the current leadership of the Supreme Administrative Court and its Chancellery is unambiguous. The exercise of judicial power in accordance with constitutional principles and relevant legal provisions belongs exclusively to judges, and the final decision-making must therefore be (his/her) human activity. AI must not be the autonomous author of a decision on the merits or the bearer of responsibility for the decision. However, this does not exclude the possible use of AI in supporting and assisting activities of judges and court officials, as long as the use of AI is in accordance with regulatory requirements and brings about the brings the desired efficiency to the respective activities in application practice. 

From the perspective of European regulation, the most relevant for the use of AI is Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence and amending Regulations (EC) No 300/2008, (EU) No 167/2013, (EU) No 168/2013, (EU) 2018/858, (EU) 2018/1139 and (EU) 2019/2144 and Directives 2014/90/EU, (EU) 2016/797 and (EU) 2020/1828 (Artificial Intelligence Act) (hereinafter ‘AI Act’). The main philosophy of the AI ​​Act, directly applicable in all EU Member States, consists of a risk-based regulatory approach – more precisely, the higher the scope and intensity of the risks of an AI system to health, safety and fundamental rights protected by the EU Charter of Fundamental Rights, the more requirements need to be met when developing and deploying AI. In the context of deploying AI in the judiciary, it is particularly necessary to consider whether, if the AI falls within the established definition of an AI system, it will not be considered a high-risk system within the meaning of Article 6(2) in conjunction with Annex III of the AI ​​Act, i.e. an AI system intended to assist a judicial authority in researching and interpreting facts and the law and in applying the law to a concrete set of facts. It is also necessary to bear in mind the possible exceptions to high-risk systems provided for in Article 6(3) of the AI ​​Act (for example, if the AI ​​system is intended to improve the outcome of a previously completed human activity). 

As for the actual use of high-risk AI systems, this generally requires the fulfillment of several obligations listed in Art. 8 et seq. of the AI ​​Act (among others, a risk management system, transparency, development and maintenance of technical documentation, human oversight). In practice, this means that the deployment of such a system is not just a technical implementation of the AI ​​tool, but a complex process including risk management, quality control and origin of input data, their secure anonymization, ongoing verification of outputs, mechanisms of transparency and explainability, as well as the setting up of effective human oversight at multiple levels. The design must also include the ability to work with data in different modes – from anonymized inputs to controlled deanonymization of outputs in justified cases. The system must also ensure full auditability of processing, retrospective control of results and identification of errors or inaccuracies. Such an approach represents a significantly higher technical and procedural complexity compared to conventional information systems and requires continuous monitoring and evaluation of the system’s behavior after its deployment. 

For the sake of completeness, it is appropriate to mention here that further guidelines on high-risk AI systems are—assuming that the relevant provisions enter into force on August 2, 2026—expected from the European Commission in the course of this year. However, at the European level, the amendment process of the AI ​​Act is underway in the form of the so-called Digital Omnibus, the result of which may also be a postponement of the entry into force of the provisions concerning high-risk AI systems.

Our own AI system – for what specific purposes and why does the Chancellery of the Supreme Administrative Court choose an internal (on-premise) AI solution? 

The Chancellery of the Supreme Administrative Court is developing its own AI system for certain activities, built internally as an on-premise solution (as recommended for public sector bodies), based on open-source large language models, which it plans to deploy into the information environment gradually. The identification of activities whose performance AI can make more efficient within the institution, together with the setting of phasing, is carried out primarily on the basis of information obtained from an internal questionnaire, as well as from other internal inputs. 

We perceive the added value of using AI primarily in more sophisticated searches for case law or other relevant sources (literature, legislation), identifying contradictions in the case law line, stylistic and grammatical control of previously created human outputs, language translations of texts, or initial control of compliance with the conditions of judicial proceedings. As is evident from the listed items, we are targeting the AI system to function as an assistive and supportive system, under no circumstances as a replacement for the lawful judge. Final decision-making, legal reasoning and evaluation of evidence will remain (and must remain) exclusively human activity. 

When developing AI and deploying it into our internal environment, in addition to the concepts outlined above, we respect the following principles: 

• independence and impartiality of the judge, 

• protection of privacy, personal data and sensitive data, 

• quality data integrated from verified and official sources, 

• identification of risks and determination of appropriate measures, 

• transparency, explainability and auditability, 

• human oversight in the development, deployment and use of AI. 

It goes without saying that even before the AI ​​is launched, along with drawing up the technical and other related documentation, an internal directive on the use of AI will be adopted. This will complement the existing legislative framework, which already regulate the use of a licensed version of a translator by legal analysts. 

Alongside the preparation of the technical solution itself, we place a strong emphasis on building AI literacy, as one of the requirements of the AI ​​Act, and for this purpose we support the participation of judges and employees not only in training with external providers, but we also organize lectures on the premises of the institution ourselves. In the future, we plan to provide AI training primarily through members of the internal AI team. 

So why does the Chancellery of the Supreme Administrative Court choose an internal (on-premise) AI solution? First of all, the use of freely available AI tools not only in the judicial environment runs into fundamental limitations in terms of trustworthiness, security, and control over the processed data as well as the outputs themselves. Public and freely available AI (chatbots) are based on general models trained on vast amounts of data from the Internet, while their outputs represent probabilistically generated text that may not be based on authoritative legal sources relevant to the legal order of the Slovak Republic and the EU. Even with correctly formulated input, inaccuracies, distortions or the creation of non-existent legal references may occur. 

However, from the point of view of the Chancellery of the Supreme Administrative Court, the question of control is crucial. With external AI services, it is not possible to fully guarantee where and how data is processed, who has access to it, how long it is stored, or whether it is subsequently used for further training of models. Such a processing model is not compatible with the requirements for the protection of sensitive data, the confidentiality of judicial information and the independence of the judiciary. 

For these reasons as well, the Chancellery of the Supreme Administrative Court chose to build its own AI solution in on-premise mode, i.e. within its own IT infrastructure. This approach allows all data to remain exclusively in the controlled environment of the institution and not be provided to third parties. At the same time, it allows for the precise definition of what data the model works with, what resources it uses and how it generates its outputs. Crucially, the internal solution also provides a high level of auditability and predictability of the system’s behaviour. The model can be continuously tested, evaluated and modified so that its outputs meet the requirements of legal practice and that there are no undesirable deviations in its behavior. Unlike external services, there are no uncontrolled changes to the model caused by updates from the provider. 

The chosen approach thus allows AI to build a precisely defined and controlled “workspace” in which it works only with verified and relevant data, respects the specifics of Slovak and European jurisdictions, and at the same time minimizes the risk of inaccurate or misleading outputs. The reason why the Chancellery of the Supreme Administrative Court does not use freely available chatbots is therefore not technological resistance, but the requirement for trustworthy, safe and fully controlled use of AI in the judicial environment.