Artificial Intelligence in Constitutional Adjudication: Implementation Practices, Limits of Use and Standards of Judicial Oversight

Ian Bernaziuk Doctor of Law, Professor Honoured Lawyer of Ukraine Judge of the Cassation Administrative Court within the Supreme Court Scientific Expert, CEPEJ-GT-CYBERJUST, Council of Europe ORCID: 0000-0002-2353-4836 Scopus Author ID: 57215524197 Web of Science ResearcherID: E-8029-2017

ARTIFICIAL INTELLIGENCE IN CONSTITUTIONAL ADJUDICATION: IMPLEMENTATION

PRACTICES, LIMITS OF USE AND STANDARDS OF JUDICIAL OVERSIGHT

An Author's Analytical Review as of 2 October 2026 Version 1.0 This is the author's English translation of the Ukrainian original, “Штучний інтелект у конституційному судочинстві: практика впровадження, межі використання та стандарти суддівського контролю”, first published on 2 October 2026 at constitutionalist.com.ua [URL: https://constitutionalist.com.ua/shtuchnyj-intelekt-u- konstytutsijnomu-sudochynstvi-praktyka-vprovadzhennia-mezhi-vykorystannia-ta- standarty-suddivskoho-kontroliu]. In the event of any discrepancy, the Ukrainian text shall prevail.

Scope of the Review This review examines publicly available information on the use, development or officially planned deployment of artificial intelligence systems in the activities of the constitutional courts of Türkiye, Peru, Colombia, the Czech Republic, the Slovak Republic, the Republic of Korea, Spain and Indonesia, as well as materials concerning institutional discussion of AI use involving the Constitutional Court of the Republic of Moldova. The experience of the Federal Supreme Court of Brazil (Supremo Tribunal Federal, STF), which exercises constitutional jurisdiction alongside its other powers, is considered separately.

The review is based primarily on judicial decisions, internal regulatory instruments, technical requirements for system development, official court communications and Council of Europe documents. Secondary academic sources [10; 24–27] are used only to reconstruct the development process of particular systems, analyse the risks associated with specific functions, and empirically assess the capabilities of large language models in constitutional interpretation; the actual status of court systems is established on the basis of official materials.

This is an independent scholarly study intended for professional discussion. It is not an official review, interpretation or position of the Supreme Court, the Council of Europe, CEPEJ or any of the courts whose activities are analysed.

Abstract This review examines the practice of using artificial intelligence in the activities of constitutional courts and identifies the functional limits of its application. It finds that relevant systems are used or developed to inform users, search case law, classify and group cases, prepare summaries of materials, generate drafts of particular parts of judicial documents, and assist in reviewing the admissibility of constitutional complaints.

The review proposes classifying such forms of use according to the nature of the function delegated to the system and the degree of its potential influence on the judicial outcome.

Six levels of AI use are identified, ranging from information retrieval to assistance in procedural legal assessment; the review also identifies functions that must not be delegated to AI and cross-cutting risks affecting access to constitutional justice and independent judicial judgment.

On the basis of Judgment T-323/2024 of the Constitutional Court of Colombia, the Protocol of the Constitutional Court of Peru, the CEPEJ Guidelines and Regulation (EU)

2024/1689, the review argues that the assessment of evidence, final interpretation of law, formulation of reasons and rendering of a judicial decision cannot be replaced by the output of an artificial intelligence system. In this context, judicial oversight means a genuine ability to verify sources, assess the completeness of the material, reject the output proposed by the system and provide independent reasons for the decision.

The review formulates requirements for the proper use of such systems: a clearly defined purpose, prior risk assessment, phased implementation, use of reliable legal sources, traceability, protection of judicial data, auditing, and preservation of the judge's ultimate responsibility for the decision.

Keywords: artificial intelligence; constitutional court; constitutional adjudication; constitutional complaint; judicial independence; right to a fair trial; judicial oversight; case law; CEPEJ; HUDERIA.

Normative Benchmarks and Limits of Comparison The CEPEJ Guidelines on the Use of Generative Artificial Intelligence for Courts, adopted in December 2025, are a specialised Council of Europe instrument concerning the use of such systems in the justice sector [1]. Under paragraph 55, decision-making, legal reasoning and the assessment of evidence should not be delegated to generative artificial intelligence. Paragraph 56 places ultimate and exclusive responsibility for the decision on the judge; paragraphs 57–58 provide that system recommendations are non-binding and protect independent judicial judgment from technological influence [1].

Paragraph 26 of the Guidelines is of particular importance for constitutional jurisdiction: the analysis of constitutionality or compliance with international treaties “requires a level of axiological interpretation that exceeds the current capabilities of AI. Even if this were not the case in the future, the issue can only be left to human judgment, since it affects the way in which citizens have decided to govern themselves” [1, para. 26].

A further distinctive feature of constitutional adjudication is that a technological error may affect not only the disposition of an individual case, but also a person's access to constitutional justice, the formation of the court's constitutional-law position, and the independence of each judge's judgment in collegiate deliberation. The assessment of AI systems must therefore take account not only of the proximity of their function to the final decision, but also of their potential impact on these three elements.

Similar risks have been identified in recent studies of large language models in constitutional interpretation. Coan and Surden show that the outputs of large language models depend significantly on the formulation of the prompt and the counterarguments supplied, and that the use of AI does not eliminate normative judgment but may merely relocate it across different stages of decision-making [24]. An empirical study by Bignotti and Camassa, based on 17 decisions of the Italian Constitutional Court from 1975 to 2023 concerning bioethical issues, likewise found that GPT-4 was capable of summarising and analysing legal materials, while also displaying a tendency towards progressive constitutional interpretations and insufficient consideration of competing values, thereby

confirming the need for human oversight [25]. The Second CEPEJ Report of 2026 recorded 176 cyberjustice and AI tools; as of the preparation of the report, there had been “no reports of fully automated decision-making processes without any human supervision” [2, para. 18, fn. 15].

A distinct instrument for assessing the risks and impacts of AI systems on human rights, democracy and the rule of law is the Council of Europe HUDERIA Methodology [3]. The Methodology was adopted by the Council of Europe Committee on Artificial Intelligence on 28 November 2024 and approved by the Committee of Ministers on 26 February 2025.

The Context-Based Risk Analysis (COBRA) resources, which form part of the HUDERIA Model, were adopted by the Committee on Artificial Intelligence on 5 November 2025; the HUDERIA Model was approved by the Committee of Ministers on 25 February 2026. The consolidated edition of the Methodology and Model was published in 2026 [3].

Regulation (EU) 2024/1689 on artificial intelligence provides an additional normative benchmark [4]. Point 8(a) of Annex III classifies as high-risk those systems intended to be used by a judicial authority or on its behalf to assist a judicial authority in researching and interpreting facts and the law and in applying the law to a concrete set of facts.

Article 6(3) of the Regulation provides that a system referred to in Annex III is not considered high-risk where it does not pose a significant risk of harm to the health, safety or fundamental rights of natural persons, including by not materially influencing the

outcome of decision-making, and one of four conditions is fulfilled: the system performs a narrow procedural task; improves the result of a previously completed human activity; detects decision-making patterns or deviations from prior decision-making patterns and is not meant to replace or influence a previously completed human assessment without proper human review; or performs a preparatory task relevant to an assessment referred to in Annex III. An Annex III system that performs profiling of natural persons is always considered high-risk [4, Art. 6(3)].

Recital 61 of the Regulation links high risk to systems intended to assist a judicial authority in researching and interpreting facts and the law or in applying the law to a concrete set of facts. At the same time, purely ancillary administrative activities that do not affect the actual administration of justice in individual cases fall outside that category, including anonymisation or pseudonymisation of judicial decisions, internal communication and administrative operations [4, recital 61].

These provisions are used as normative benchmarks but do not establish any automatic correspondence between the levels of the classification proposed in this review and the risk categories under the Regulation. As EU law, the Regulation is directly relevant primarily to EU Member States; in relation to the other jurisdictions examined, it is used for comparative purposes. The comparison also does not imply equivalence between the competences of the courts examined, in particular the Brazilian STF and specialised constitutional courts.

Source Base, Selection Criteria and Evidentiary Status of Sources The source base was compiled through a search of publicly available official materials of constitutional courts and international organisations. Materials were included where they made it possible to establish reliably at least one of the following circumstances: actual use of a system by the relevant court; officially confirmed development of the system; a specifically identified functional purpose; or regulation of AI use by the constitutional court itself.

A system in operational judicial use, a pilot project, a tool under development, an officially announced plan for future use, and a training or consultation event have different evidentiary status and are distinguished throughout this review.

Within the scope of the search conducted, no sufficient publicly available official evidence was identified to include in the main analysis any in-house AI systems of the Federal Constitutional Court of Germany, the Constitutional Council of France or the Constitutional Court of Italy. This is not a statement that those institutions lack any internal technological tools; it means only that the publicly available materials are insufficient to describe such tools properly under the criteria of this review.

Kazakhstan is likewise not included in the principal source base. Some public reports refer to the development of an AI portal for the Constitutional Court, but the available material is insufficient to establish reliably its functions, state of implementation and limits of use.

The review does not purport to provide an exhaustive global list of courts exercising constitutional review. The analysis includes those examples for which open sources make it possible to establish with sufficient precision the function of the system, the stage of its implementation and its connection with the exercise of constitutional jurisdiction.

Access to Information on Constitutional Adjudication The Constitutional Court of Korea has officially operated the “헌재톡 ” chatbot and an intelligent integrated search system since 1 January 2023; the official notice was published on 19 January 2023 [18]. The chatbot provides round-the-clock information on constitutional adjudication, the requirements for a constitutional complaint and the procedure for its consideration, while the intelligent search function allows queries to be formulated using legal terms, ordinary words and full sentences.

Since 1 July 2026, the Constitutional Court of the Slovak Republic has provided access to the JURO digital assistant [17]. It works with published decisions and materials from the Court's official website. The Court cautions that JURO does not provide legal advice, binding interpretations of law or an official position of the Constitutional Court. Its responses may be incomplete or inaccurate and must be verified against primary court materials [17].

On 23 February 2026, after several weeks of pilot operation, the Constitutional Court of the Czech Republic announced an updated version of an AI-based chatbot for its official website and the NALUS database [16]. Its purpose is to facilitate access to the substantial body of constitutional case law and enable information retrieval in several languages.

This experience also has a direct Ukrainian dimension. On 14 July 2026, the Constitutional Court of Ukraine reported that, following study visits by its judges and Secretariat staff to the Constitutional Court of the Czech Republic in February 2026, the Czech Court launched a Ukrainian-language AI assistant in the NALUS database. According to the Constitutional Court of Ukraine, this facilitated access for Ukrainian judges, lawyers and researchers to Czech constitutional case law and made it a practical instrument of comparative legal analysis in the work of the Constitutional Court of Ukraine [28].

In these examples, the system does not directly participate in deciding a particular case, but it influences users' access to case law through document selection and the generation of answers. Under the classification proposed in this review, such influence falls within the lowest level (level 1).

Search, Classification and Grouping of Cases

The PretorIA system of the Constitutional Court of Colombia was created to process a large volume of tutela cases (constitutional proceedings for the protection of fundamental rights) and to support procedures for selecting cases for possible review [10].

Its subsequent actual use is confirmed by the Constitutional Court itself in Judgment T- 127/2026 [11]. In paragraph 154, the Court expressly describes PretorIA as an institutional tool for documentary and statistical analysis and explains its use for classification, filtering and searching within a large volume of cases. Paragraph 156 describes the subsequent manual verification of the results: of 181 results, 141 were found after individual review to be clearly related to the issue under examination [11, paras. 154, 156].

In Brazil, the STF introduced the VICTOR system in 2018. Official Court materials identify among its functions the conversion of images into text, separation of an electronic case file into individual procedural documents, classification of documents, and identification of repercussão geral themes (general repercussion, a threshold requirement for the admissibility of extraordinary appeals) [13].

A subsequent tool, VitórIA, was integrated into STF-Digital on 17 May 2023 pursuant to STF Resolution No. 800/2023 [14]. Its purpose is to group cases thematically by similarity and facilitate the identification of legal issues recurring across a significant number of proceedings.

Such systems do not formulate the final legal conclusion, but they determine the body of information with which the court subsequently works. The classification criteria, completeness of results, grounds for assigning a case to a particular group, and the ability to identify materials not selected by the system therefore require verification.

Academic analysis of PretorIA shows that case selection is not a purely technical operation: algorithmic classification may lead not only to erroneous inclusion but also to the omission of relevant cases, with the consequence that the court may fail to receive information about a new or as yet unrecognised type of social conflict [26]. Research into the development of PretorIA also indicates that the system's actual functional capabilities became considerably more limited than the original technological expectations, in particular because of risks of bias, manipulation and weakening of judicial oversight [27].

Summarisation of Materials and Preparation of Draft Documents On 29 January 2026, the Constitutional Court of Peru reported the use of generative artificial intelligence to prepare summaries of materials before their distribution to chambers and the plenary, to generate the antecedentes section (a statement of the prior circumstances and case materials), to create a specialised database for search, and for subsequent work connected with the preparation of draft judicial decisions [9].

On 16 December 2024, the Brazilian STF launched MARIA – Módulo de Apoio para Redação com Inteligênc ia Artificial, the Court's first in house generative artificial intelligence tool [15]. The STF's official communication identifies three initial areas of use: preparation of

draft ementas (concise official summaries of a judicial decision) based on the content of the Justice's vote (voto); preparation of procedural reports in extraordinary-appeal proceedings; and preliminary analysis of applications in cases of the Reclamação class (a special procedure for bringing a complaint before the STF) [15].

The last of these functions involves analysing the initial application and providing structured answers to questions used in the preliminary examination of the case. This goes beyond simple linguistic editing, but the available official description does not provide a sufficient basis for equating the function with a legal assessment of the admissibility of the application.

The preparation of a summary or a draft document does not automatically determine the outcome of the case. At the same time, an incomplete statement of facts, omission of an argument or erroneous selection of case law may affect the subsequent analysis. The completeness of the material, accurate representation of the parties' positions, appropriateness of the sources used and absence of distortion of the substance of the case must therefore be verified.

In a collegiate court, an additional risk arises where a system-generated summary becomes a common informational basis for several judges. In that situation, each judge must retain a genuine ability to consult the full materials and verify the account produced by the system.

Admissibility Review of Constitutional Complaints: The Experience of Türkiye Among the projects examined, the project of the Constitutional Court of Türkiye comes closest to direct procedural legal assessment.

The technical specifications provide for converting case materials into machine-readable text, semantic analysis, searching for similar applications and previous decisions, preparation of a concise summary of the application, and drafts containing the sections “Başvurunun Özeti” (“Summary of the Application”) and “Hukuki Değerlendirme” / “Değerlendirme” (“Legal Assessment” / “Assessment”). The stated purpose is to assist in preparing draft admissibility decisions; the system is described as a decision-support tool [5].

As of 20 May 2026, the technical document recorded 737,860 individual applications lodged since 23 September 2012, of which 102,915 remained pending. The training data are intended to include 295,000 inadmissibility decisions of the Commissions and 17,015 decisions of the Court's Sections [5].

The substantial predominance of inadmissibility decisions in the training corpus does not in itself establish bias in the system, but it requires separate examination of whether the structure of the training data creates a risk of disproportionate reproduction of inadmissibility patterns in the system's outputs. Such scrutiny is particularly important for a system used at the stage that determines access to constitutional justice.

The technical specifications also provide for identification of similar applications and previous decisions, citation of at least three decisions of the Constitutional Court, measures to reduce the risk of fabricated information, and classification of applications according to groups of rights defined by the Court [5].

In a speech delivered on 28 April 2026, the President of the Constitutional Court, Kadir Özkaya, stated that the Court planned to use AI to read, summarise and classify individual applications; the system's outputs would have no independent legal effect and would be reviewed by judge-rapporteurs [6].

At the symposium marking the fourteenth anniversary of the introduction of the individual application, held on 23 September 2026, the President of the Court stated that the system designed to assist with the admissibility assessment of individual applications was nearing completion. The official text separately refers to information and data security, the central importance of the human element in the administration of justice, and the preservation of human responsibility for judicial decisions [7].

Regulation of AI Use: The Experience of Peru On 16 September 2025, the Plenary of the Constitutional Court of Peru approved a dedicated Protocol for the Use of Artificial Intelligence; its approval was formalised by Administrative Resolution No. 115-2025-P/TC of 19 September 2025 [8].

Article 6 of the Protocol permits assistance in preparing reports and drafts based on case materials, subject to mandatory supervision and review by the responsible person; automated search and classification of proceedings; search and classification of relevant case law; and summarisation and systematisation of previous judicial decisions [8, Art. 6].

Article 7 establishes the limits of such use. Prohibited forms include substituting AI for the legal analysis and assessment of evidence reserved exclusively to constitutional judges, using the system to influence the independence or impartiality of the court, and unlawfully disclosing confidential information [8, Art. 7].

Articles 8–11 regulate documentation of the system's operation, access to information, notification of the public about AI use in proceedings, protection of personal data, interoperability with other systems, staff training and internal auditing [8, Arts. 8–11].

The Protocol was approved before the Court's official communication of 29 January 2026 concerning the practical use of generative AI tools [8; 9]; in other words, dedicated regulatory rules preceded publicly confirmed practical use.

Limits on the Delegation of Judicial Functions One of the most detailed judicial approaches to the limits of generative artificial intelligence use was set out by the Constitutional Court of Colombia in Judgment T- 323/2024 of 2 August 2024, when reviewing the use of ChatGPT by a lower-court judge [12]. The standards formulated by the Court are relevant not as an example of AI use by

the Constitutional Court itself, but as a constitutional-law assessment of the permissible limits of such use in the administration of justice.

In paragraph 296, the Court stated that AI may be used in the administration of justice provided that those adjudicative functions requiring independent judicial judgment remain with a human decision-maker, including interpretation of facts and evidence, reasoning and decision-making [12, para. 296].

At the same time, the Court emphasised that the absence of substitution for the judge does not remove the requirements of transparency and accountability: in the particular case they had not been fully complied with, and some of the information supplied by ChatGPT proved inaccurate [12, paras. 360, 364–366]. It is therefore necessary to establish not only when the system was used, but also which intellectual operations the judge performed independently, how the output was verified and the extent to which the judge relied on it.

The Peruvian Protocol draws the corresponding boundary through regulation, reserving legal analysis and assessment of evidence to constitutional judges [8, Art. 7]. CEPEJ formulates a general standard: judicial decision-making, legal reasoning and the assessment of evidence should not be delegated to generative AI, while ultimate and exclusive responsibility for the judicial decision remains with the judge [1, paras. 55–58].

The mere fact that the final document is signed by a judge does not guarantee adequate judicial oversight. The judge must in fact be able independently to examine the materials, assess the evidence and legal sources, reject the proposed output and formulate his or her own reasoning.

In this review, the term judicial oversight does not denote a separate procedural form of judicial review. It means substantive oversight by the judge over the use of AI: access to the primary materials and legal sources, verification of the completeness of the information selected, the ability to modify or reject the system's output, the capacity independently to explain the reasons for the decision, and preservation of the judge's ultimate responsibility for its content.

This approach is consistent with Opinion No. 26 (2023) of the Consultative Council of European Judges: “Decision-making must, explicitly and implicitly, only be carried out by judges. It cannot be delegated to or carried out by or through technology” [23, para. 93(iii)].

Projects at the Development and Institutional Preparation Stage In a decision of 23 April 2026 issued in response to a request for public information, the General Secretariat of the Constitutional Court of Spain stated that, for the period beginning on 1 January 2023, the Court had one project using artificial intelligence technologies and that the project was not yet in operational use [19].

The project concerns the digital processing of approximately 3,000 orders issued between 1987 and 2000, automated extraction of relevant information, and structured incorporation of that information into the case-law database [19]. The decision does not identify any other AI projects or systems for the period covered by the request.

The project therefore primarily concerns the information infrastructure of case law: the system works with an archival body of decisions and is not described as a tool for analysing a particular pending case.

The Constitutional Court of the Republic of Indonesia included in its 2027 work plan and budget the transformation of a “digital Constitutional Court”, integration of judicial and administrative management systems, and expanded use of AI to support judicial services and institutional administration [20]. The official material confirms the planned direction of development but does not establish that the relevant judicial system has entered operational use.

On 22 May 2026, CEPEJ held a dedicated event in cooperation with the Constitutional Court of the Republic of Moldova. Judges and staff of the Constitutional Court participated; the discussion addressed the opportunities and risks of AI, judicial independence, transparency, generative systems, meaningful judicial oversight, and the human dimension of judicial deliberation in constitutional adjudication [22]. This material confirms institutional discussion of the opportunities, risks and safeguards associated with AI use, but does not establish a decision to deploy a specific system.

Ukraine's national framework governing the Constitutional Court likewise makes it possible to distinguish auxiliary technological operations from judicial legal judgment proper. Under part four of Article 59 of the Law of Ukraine “On the Constitutional Court of Ukraine”, the judge-rapporteur examines the issues raised in the application and prepares materials for consideration by a Panel, Senate or the Grand Chamber [29]. The Rules of Procedure of the Constitutional Court of Ukraine provide for the Secretariat to prepare a preliminary opinion and analytical, reference and informational materials, whereas final discussion of a draft act, submission of proposals and amendments, and voting are carried

out by the judges of the relevant body of the Court [30]. Under this allocation of responsibilities, auxiliary use of AI for information retrieval, systematisation or preparation of materials may be assessed within the existing distribution of functions between the Secretariat and the judge-rapporteur. By contrast, use of AI to conduct an independent legal assessment of the admissibility of a constitutional complaint would require separate normative definition of its limits and safeguards.

This allocation of responsibilities provides a normative benchmark for defining the limits of technology use at auxiliary stages, provided that such use does not substitute for the judge's independent examination of the materials, legal assessment of admissibility, collegiate deliberation or voting.

This approach is also consistent with publicly expressed positions of judges of the Constitutional Court of Ukraine concerning the use of AI. During an official lecture in February 2024, Judges Viktor Horodovenko and Halyna Yurovska of the Constitutional Court, discussing CCJE Opinion No. 26, emphasised that AI should support rather than replace judges, that ultimate responsibility for the decision remains with the human decision-maker, and that technology must not be used to delegate judicial decision- making [31].

Levels of Artificial Intelligence Use in Constitutional Adjudication The proposed classification is based on the nature of the function delegated to the system and the typical proximity of its output to the legal determination of the case; the actual influence in a particular proceeding may differ depending on how the system is used and on automation bias. The classification is not a scale of technological maturity or lawfulness. A formally lower level likewise does not mean the absence of substantial influence: a selection or classification system may determine the body of information that actually reaches judicial analysis, and therefore particular oversight is required not only over erroneous inclusion of cases but also over the failure to identify cases relevant to

subsequent analysis [26]. Unlike the functional grouping of tools in the CEPEJ Resource Centre and the normative risk classification under Regulation (EU) 2024/1689, the proposed model arranges forms of AI use according to their proximity to independent judicial legal judgment and separately identifies judicial functions that must not be delegated to AI [2; 4].

As the influence on research into facts and law, assessment of materials, a case's access to further consideration, and formulation of reasons increases, the requirements of judicial oversight also become more stringent.

Levels of Use of Artificial Intelligence Systems in Judicial Activity and Minimum Safeguards for Their Application Level Function Nature of impact Examples Minimum safeguards Defined body of sources; Korea [18]; Informing users and Indirect influence links to decisions; disclaimer Slovakia [17]; 1 searching published case through access to that the response is Czech Republic law information informational; ability to [16] verify Understandable Classification and Influence on the classification criteria; filtering of cases body of material PretorIA [10; 11]; 2 completeness checks; according to defined subsequently VICTOR [13] possibility of manual criteria examined verification

Level Function Nature of impact Examples Minimum safeguards Influence on Audit of similarity criteria; Semantic identification of identification of ability to verify group 3 similar cases and recurring legal VitórIA [14] composition and omitted thematic grouping issues and relevant cases case law Peru – summarisation of materials [9]; Formation of a MARIA - Summarisation of structured preparation of Comparison with the materials, preparation of preliminary procedural complete case file; 4 reference notes and understanding of reports [15]; traceability of sources; auxiliary texts facts, arguments Türkiye - completeness review and case law “Başvurunun Özeti” (summary of the application) [5] Peru -

preparation of Potential influence antecedentes Mandatory legal review; Preparation of drafts of on the structure of [9]; MARIA – ability to fully rework the 5 parts of a judicial or reasoning and draft ementas text; documentation of procedural document content of the future [15]; Türkiye – system use act draft admissibility decisions [5] Enhanced judicial oversight; Direct assistance in Turkish explainability; advisory the legal Assistance in procedural admissibility- nature of the output; ability 6 assessment of a legal assessment review project of the judge to reject it; particular application [5–7] audit; ultimate responsibility as to its admissibility of the judge Independent assessment No confirmed

of evidence, final legal instances of These functions are Non- interpretation and Determines the autonomous performed by the judge; AI delegabl application of law to the outcome of the performance of may be used only in an e facts of the case, administration of these functions auxiliary capacity while function formulation of reasons justice were identified preserving judicial oversight s and rendering of the in the reviewed [1; 8; 12] decision source base Filtering or classifying a body of cases does not, by itself, constitute level six; the determining factor is the use of the system for legal assessment of the admissibility of a particular application.

At the same time, influence on whether a case proceeds to further consideration is a cross- cutting risk factor: if a system at any level in fact affects the selection of cases or their admission to further consideration, enhanced safeguards are required irrespective of where the relevant function falls within the classification.

The distinction between levels is determined by the substance of the operation. A single system may simultaneously encompass several levels depending on the function performed; for example, it may provide information retrieval while also participating in the formulation of a legal assessment.

This functional approach is consistent with Regulation (EU) 2024/1689: Article 6(3), Annex III, point 8(a), and recital 61 distinguish preparatory and administrative functions from systems that participate in researching and interpreting facts and the law [4]. At the same time, the proposed classification does not reproduce the Regulation's categories and does not replace an individual assessment of a specific system; its purpose is to determine how closely a technological function approaches the sphere of independent judicial legal judgment.

Level six requires the most stringent safeguards among the forms of use identified in this classification. A recommendation on admissibility may directly affect whether a constitutional complaint proceeds to further consideration and, consequently, a person's access to constitutional justice.

Conditions for Proper Use Use of a system should be preceded by identification of a specific functional need. Paragraphs 13–15 of the CEPEJ Guidelines provide for first identifying the problem, then assessing possible technological tools, and only thereafter selecting the appropriate system [1].

Heavy caseloads, difficulty of searching a large body of case law, repetitive technical operations and the need for digital processing of historical documents are different tasks and may require different technological solutions.

Before a system is placed into operation, its potential impact on the right to a fair trial, access to justice, privacy, equality, judicial independence and effective legal protection must be assessed. HUDERIA provides for contextual risk analysis, stakeholder engagement, impact assessment, identification of mitigation measures and subsequent review [3].

The legal sources with which the system works must be reliable and verifiable. Paragraphs 52–54 of the CEPEJ Guidelines associate fabricated legal provisions or judicial citations with risks to legal certainty and provide for automatic verification of legal citations, exclusive use of certified official legal sources, and flagging of information that cannot be verified against authoritative legal databases [1].

Traceability makes it possible to establish which sources the system processed, what output it generated, who reviewed it and the extent to which the output was used.

Paragraph 61 of the CEPEJ Guidelines provides for documenting the use of generative AI in the preparation of a judicial decision and for the parties' right to know which elements of the decision were prepared with the assistance of such a system [1].

Judicial data require a dedicated protection regime. Uncontrolled transfer of case materials to an external provider, their use for further training of a general-purpose model, or lack of control over where the information is stored may create risks to confidentiality and the independence of the judicial system. CEPEJ gives preference to specialised and controlled solutions over uncontrolled use of publicly available commercial tools [1, paras.

27–32]. Judicial oversight includes the ability to assess the system's output critically. Paragraphs 55–60 of the CEPEJ Guidelines require the preservation of autonomous judicial judgment and provide for measures to prevent automation bias [1].

Related practice in Ukrainian administrative adjudication confirms a broader principle: automation may change the technological manner in which a legally significant function is performed, but does not transfer legal responsibility to the automated system; the decisive factors remain the normative allocation of powers, verifiability of the output, and responsibility of the person or authority that gives the output legal effect [32]. The auxiliary role of AI in searching, analysing and systematising case law has also previously been substantiated by the author in the context of ensuring the consistency of case law [33].

Training of judges and court staff should ensure appropriate competence in the use of AI: knowledge of the system's limitations, typical sources of error, sensitivity of outputs to prompt formulation, methods for verifying sources and comparing alternative outputs, and the circumstances in which use of a particular tool is inappropriate or impermissible [24; 25; 31].

Conclusion The practice examined demonstrates a broad range of artificial intelligence functions in constitutional adjudication: from searching, classifying and processing large volumes of cases to summarising materials, preparing parts of judicial documents and assisting in the assessment of the admissibility of constitutional complaints. No instance was identified in the materials reviewed in which a system autonomously rendered a final decision in constitutional proceedings.

For legal assessment, the decisive factors are the function of the system and the degree of its influence on the outcome of the proceedings. The proposed six levels of AI use and the separate sphere of judicial functions that must not be delegated to AI are based on those criteria.

Three aspects are of particular importance for constitutional adjudication: the influence of AI on whether a case proceeds to further consideration, the impermissibility of substituting AI for axiological constitutional interpretation, and preservation of the independent judgment of each judge in a collegiate process. These risks may arise at different levels of the proposed classification and therefore require separate assessment irrespective of the system's technical function.

AI should be introduced into constitutional courts in stages, using specialised tools for searching, classification, grouping, summarisation and preparation of auxiliary documents. Use of AI for procedural legal assessment requires enhanced safeguards and must not reduce a judicial decision to formal endorsement of an output generated by a system.

For the Constitutional Court of Ukraine, the issue of AI use is no longer purely abstract. Institutional cooperation with the Constitutional Court of the Czech Republic has given Ukrainian users access to the Ukrainian-language NALUS AI assistant [28], while judges of the Constitutional Court of Ukraine have publicly discussed the limits of assistive technologies [31]. The proposed classification may serve as a framework for assessing future AI tools in the Constitutional Court of Ukraine: the closer the system's function comes to the legal assessment of a particular constitutional complaint, formulation of reasons or the outcome of proceedings, the higher the requirements for verifiability,

transparency and judicial oversight should be. Final constitutional interpretation, reasoning and decision-making must remain the product of independent and collegiate judicial judgment [29; 30].

References All online sources last accessed: 2 October 2026.

1. European Commission for the Efficiency of Justice (CEPEJ). Guidelines on the Use of

Generative Artificial Intelligence for Courts. CEPEJ(2025)18Final. Adopted at the 45th Plenary Meeting of the CEPEJ, Strasbourg, 4–5 December 2025.

URL: https://rm.coe.int/cepej-2025-18final-en-draft-guidelines-on-the-use-of-generative- ai-for/48802a4ad1

2. European Commission for the Efficiency of Justice (CEPEJ). 2nd Report on the Use of

Artificial Intelligence (AI) in the Judiciary Based on the Information Contained in the CEPEJ Resource Centre on Cyberjustice and AI. CEPEJ(2026)7. Adopted at the 46th Plenary Meeting, 18–19 June 2026.

URL: https://rm.coe.int/cepej-2026-7-en-second-aiab-report-2026/48802cbf1a

3. Council of Europe. HUDERIA – Methodology and Model: Methodology for Assessing the

Risks and Impacts of Artificial Intelligence Systems from the Perspective of Human Rights,

Democracy and the Rule of Law; HUDERIA Model, including Context-Based Risk Analysis (COBRA) resources. Strasbourg, 2026.

URL: https://rm.coe.int/huderia-methodology-and-model-adopted-provisional-version- 2026/48802ac001 Committee of Ministers decision concerning the HUDERIA Methodology: https://www.coe.int/en/web/cm/-/deputies-meeting-24-26-february-2025 Information concerning the HUDERIA Model and COBRA resources: https://www.coe.int/en/web/artificial-intelligence/huderia-risk-and-impact-assessment- of-ai-systems

4. Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June

2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). OJ L, 2024/1689, 12 July 2024. Article 6; Annex III, point 8(a); recital 61.

URL: https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng

5. T.C. Anayasa Mahkemesi. Yapay Zekâ Destekli Başvuru Değerlendirme Sistemi Projesi

[AI-Supported Application Assessment System Project]. Teknik İster Dokümanı [Technical Specifications Document]. 2026. URL: https://tubitak.gov.tr/sites/default/files/2026- 06/AYM_Teknik_Ister_Dokumani_0.pdf

6. Constitutional Court of the Republic of Türkiye. Özkaya K. Opening Remarks at the

Official Ceremony Marking the 64th Anniversary of the Constitutional Court and the Swearing-in of Member Mr Şaban Kazdal. Ankara, 28 April 2026.

URL: https://www.anayasa.gov.tr/en/president/presidents-speeches/kadir- ozkaya/opening-remarks-official-ceremony-marking-the-64th-anniversary-of-the- constitutional-court-and-the-swearing-in-of-member-mr-saban-kazdal

7. Constitutional Court of the Republic of Türkiye. Symposium Held on the Occasion of the

14th Anniversary of the Introduction of Individual Application to the Constitutional Court. Symposium held 23 September 2026; English report dated 24 September 2026.

URL: https://www.anayasa.gov.tr/en/news/news-and-events/symposium-held-on-the- occasion-of-the-14th-anniversary-of-the-introduction-of-individual-application-to-the- constitutional-court

8. Tribunal Constitucional del Perú. Resolución Administrativa N.º 115-2025-P/TC.

Protocolo para el uso de la Inteligencia Artificial en el Tribunal Constitucional [Administrative Resolution No. 115-2025-P/TC. Protocol for the Use of Artificial Intelligence in the Constitutional Court]. 19 September 2025.

URL: https://intranet.sedetc.gob.pe/wp-content/uploads/2025/09/RA-115-2025-P-TC.pdf

9. Tribunal Constitucional del Perú. El vicepresidente del TC, Helder Domínguez, destacó

la utilización de la IA generativa en el TC [The Vice-President of the Constitutional Court, Helder Domínguez, highlighted the use of generative AI in the Court]. Lima, 29 January 2026.

URL: https://www.tc.gob.pe/institucional/notas-de-prensa/vicepresidente-del-tc-helder- dominguez-destaco-la-utilizacion-de-la-ia-generativa-en-el-tc

10. Saavedra V., Upegui J. C. IA e inclusión: Colombia. “PretorIA” y la automatización del

procesamiento de causas de derechos humanos [AI and Inclusion: Colombia. “PretorIA” and the Automation of Human Rights Case Processing]. Derechos Digitales; Dejusticia, March 2021.

URL: https://www.derechosdigitales.org/wp- content/uploads/CPC_informe_Colombia.pdf

11. Corte Constitucional de Colombia, Sala Sexta de Revisión. Sentencia T-127/26

[Judgment T-127/26], 8 May 2026. Judge-Rapporteur: Vladimir Fernández Andrade. Paras. 154–156. URL: https://www.corteconstitucional.gov.co/relatoria/2026/T-127-26.htm

12. Corte Constitucional de Colombia, Sala Segunda de Revisión. Sentencia T-323/24

[Judgment T-323/24]. Bogotá D.C., 2 August 2024. Paras. 296, 360, 364–366. URL: https://www.corteconstitucional.gov.co/relatoria/2024/T-323-24.htm

13. Supremo Tribunal Federal. Ministra Cármen Lúcia anuncia início de funcionamento do

Projeto Victor, de inteligência artificial [Justice Cármen Lúcia Announces the Launch of the Victor Artificial Intelligence Project]. 30 August 2018.

URL: https://portal.stf.jus.br/noticias/verNoticiaDetalhe.asp?idConteudo=388443&ori=1

14. Supremo Tribunal Federal. Ministra Rosa Weber lança robô VitórIA para agrupamento

e classificação de processos [Justice Rosa Weber Launches VitórIA for Grouping and Classification of Cases]. 17 May 2023. Resolução STF n.º 800/2023.

URL: https://portal.stf.jus.br/noticias/verNoticiaDetalhe.asp?idConteudo=507426&ori=1

15. Supremo Tribunal Federal. STF lança MARIA, ferramenta de inteligência artificial que

dará mais agilidade aos serviços do Tribunal [STF Launches MARIA, an Artificial Intelligence Tool Designed to Increase the Efficiency of Court Services]. Launch on 16 December 2024; official report published 17 December 2024.

URL: https://portal.stf.jus.br/noticias/verNoticiaDetalhe.asp?idConteudo=591257&ori=1

16. Ústavní soud České republiky. AI chatbot pro snazší vyhledávání v databázi NALUS i

na webových stránkách Ústavního soudu [AI Chatbot to Facilitate Searches in the NALUS

Database and on the Constitutional Court Website]. Press Release TZ 16/2026. 23 February 2026. URL: https://www.usoud.cz/aktualne/ai-chatbot-pro-snazsi-vyhledavani-v-databazi-nalus- i-na-webovych-strankach-ustavniho-soudu

17. Ústavný súd Slovenskej republiky. JURO – digitálny AI asistent Ústavného súdu

Slovenskej republiky [JURO – Digital AI Assistant of the Constitutional Court of the Slovak Republic]. 1 July 2026. URL: https://ustavnysud.sk/juro-digitalny-asistent

18. Constitutional Court of Korea. 민원상담 챗봇 ‘ 헌재톡 ’ 과 ‘ 지능형 통합검색 ’ 서비스

공식운영 안내 [Official Operation of the “HunjaeTalk” Consultation Chatbot and Intelligent Integrated Search Service]. Official operation since 1 January 2023; notice dated 19 January 2023.

URL: https://ccourt.go.kr/site/kor/ex/bbs/View.do?bcIdx=1003928&cbIdx=1127

19. Tribunal Constitucional de España, Secretaría General. Resolución de 23 de abril de

2026 de la Secretaría General del Tribunal Constitucional en relación con la solicitud de información formulada sobre la implantación de herramientas o proyectos que utilicen inteligencia artificial [Decision of 23 April 2026 of the General Secretariat of the Constitutional Court concerning a request for information on the implementation of tools or projects using artificial intelligence]. Madrid, 23 April 2026. Request 7/2026.

URL: https://www.tribunalconstitucional.es/es/transparencia/informacion- publica/Lists/ResolucionesDictadas/Attachments/609/Resolución%20de%20Secretaría %20General%20(censurada).pdf

20. Constitutional Court of the Republic of Indonesia. Court Reports 2026 Budget, 2027

Work Plan Before House Commission III. Jakarta, 31 August 2026. URL: https://en.mkri.id/news/details/2026-08- 31/Court_Reports_2026_Budget%2C_2027_Work_Plan_Before_House_Commission_III

21. Suchikova Y., Tsybuliak N., Teixeira da Silva J. A., Nazarovets S. GAIDeT (Generative

AI Delegation Taxonomy): A taxonomy for humans to delegate tasks to generative artificial intelligence in scientific research and publishing. Accountability in Research. 2026. Vol.

33, No. 3. Article 2544331. DOI: 10.1080/08989621.2025.2544331. URL: https://doi.org/10.1080/08989621.2025.2544331

22. European Commission for the Efficiency of Justice (CEPEJ). CEPEJ Online Workshop

on the Use of Artificial Intelligence in Justice in the Republic of Moldova. 22 May 2026. URL: https://www.coe.int/en/web/cepej/-/cepej-online-workshop-on-the-use-of-artificial- intelligence-in-justice-in-the-republic-of-moldova

23. Consultative Council of European Judges (CCJE). Opinion No. 26 (2023), “Moving

forward: the use of assistive technology in the judiciary”. CCJE(2023)3, 1 December 2023. URL: https://rm.coe.int/ccje-opinion-no-26-2023-final/1680adade7

24. Coan A., Surden H. Artificial Intelligence and Constitutional Interpretation. University

of Colorado Law Review. 2025. Vol. 96. P. 413–498. URL: https://scholar.law.colorado.edu/lawreview/vol96/iss2/4

25. Bignotti C., Camassa C. Legal Minds, Algorithmic Decisions: How LLMs Apply

Constitutional Principles in Complex Scenarios. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society. 2024. Vol. 7, No. 1. P. 120–130. DOI: 10.1609/aies.v7i1.31623.

URL: https://ojs.aaai.org/index.php/AIES/article/view/31623

26. Rueda Saiz P. Docket Selection and Judicial Responsiveness: The Use of AI in the

Colombian Constitutional Court. William & Mary Bill of Rights Journal. 2021. Vol. 30, Iss.

2. P. 419–454. URL: https://scholarship.law.wm.edu/wmborj/vol30/iss2/9

27. Ángel-Cabo N., Pardo-Rodríguez S. Imaginarios sociotécnicos en las promesas de la

tecnología para la selección de tutelas en la Corte Constitucional de Colombia [Sociotechnical Imaginaries in the Promises of Technology for the Selection of Tutela Cases in the Constitutional Court of Colombia]. In: Urueña Hernández R., Ángel-Cabo N.

(eds.). Derecho, poder y datos: aproximaciones críticas al derecho y las nuevas tecnologías. Bogotá: Universidad de los Andes, Facultad de Derecho, Ediciones Uniandes, 2024. P. 261–287. DOI: 10.51573/Andes.9789587986907.9789587986914.9. URL: https://doi.org/10.51573/Andes.9789587986907.9789587986914.9

28. Constitutional Court of Ukraine. Practical Experience and Shared Values: The

Constitutional Court of Ukraine Deepens International Cooperation [in Ukrainian]. 14 July 2026. URL: https://ccu.gov.ua/novyna/praktychnyy-dosvid-i-spilni-cinnosti- konstytuciynyy-sud-ukrayiny-poglyblyuye-mizhnarodnu

29. Law of Ukraine “On the Constitutional Court of Ukraine” of 13 July 2017 No. 2136-VIII

[in Ukrainian]. URL: https://zakon.rada.gov.ua/laws/show/2136-19#Text

30. Rules of Procedure of the Constitutional Court of Ukraine, approved by Resolution of

the Constitutional Court of Ukraine No. 1-ps/2018 of 22 February 2018 [in Ukrainian]. URL: https://zakon.rada.gov.ua/laws/show/v0001710-18#Text

31. Suddia i shtuchnyi intelekt: perevahy, vyklyky, zahrozy (chastyna 2) [The Judge and

Artificial Intelligence: Benefits, Challenges and Threats (Part 2)] [in Ukrainian]. Sudova vlada Ukrainy. 2 February 2024. URL: https://court.gov.ua/archive/1550733

32. Bernaziuk, Ian. Artificial Intelligence and Automated Systems in Public Administration

and Adjudication: Human Oversight, Reason-Giving, Verifiability and Responsibility in the Case Law of the Cassation Administrative Court within the Supreme Court: An Analytical

Review of Case Law. Version 1.0. 2026. Zenodo. DOI: 10.5281/zenodo.22274766. URL: https://doi.org/10.5281/zenodo.22274766

33. Bernaziuk, Ian. Artificial Intelligence and Its Use in Ensuring Consistency of Case Law

as a Component of Trust in the Judiciary [in Ukrainian]. Slovo Natsionalnoi shkoly suddiv Ukrainy [Word of the National School of Judges of Ukraine]. 2024. No. 4(49). P. 16–35.

DOI: 10.37566/2707-6849-2024-4(49)-2. URL: https://slovo.nsj.gov.ua/images/pdf/2024_4_49/nsj_4_49_2024.pdf Status of the Review The review is based on publicly available materials accessible to the author as of 2 October 2026. The status of each system – operational, pilot, under development or planned – is stated only to the extent directly supported by the sources used.

The analysis does not rely on non-public information, internal court materials or judicial deliberations. Comparative use of the experience of foreign courts does not imply that their competences are identical or that a solution adopted in one legal system may be transferred directly to another.

The synthesis concerning levels of AI use, limits on permissible delegation of functions and conditions of judicial oversight constitutes the author's scholarly assessment and does not represent the position of any court or international organisation. The review was prepared in the course of the author's scholarly activity; the views expressed do not concern the determination of specific cases pending before courts and do not predetermine the author's position in other judicial proceedings.

Transparency Declaration on the Use of Artificial Intelligence In preparing this review, ChatGPT (OpenAI) and Claude (Anthropic) were used. Following the GAIDeT approach [21], generative AI was assigned, under full authorial control, discrete auxiliary tasks relating to the search and systematisation of open sources, verification of bibliographic details, structural and linguistic editing, and identification of possible inaccuracies.

Generative AI was not delegated the definition of the subject of the study, the criteria for selecting sources, legal assessment of the materials, development of the proposed classification, formulation of scholarly conclusions, or final editorial decisions. All factual propositions and references identified or refined with the assistance of AI were verified by the author against primary sources. AI outputs were accepted, amended or rejected by the author; the final text and responsibility for its content remain with the author.

Suggested Citation Bernaziuk, Ian. Artificial Intelligence in Constitutional Adjudication: Implementation Practices, Limits of Use and Standards of Judicial Oversight (An Author's Analytical Review as of 2 October 2026), Version 1.0. Zenodo, 2026. DOI: 10.5281/zenodo.23103501.

© 2026 Ian Bernaziuk. Except for third-party material identified in the text and references, the author’s original content is licensed under the Creative Commons Attribution 4.0 International licence (CC BY 4.0). Third-party material remains subject to the rights and terms applicable to its original source. URL: https://creativecommons.org/licenses/by/4.0/