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

Ian Bernaziuk Doctor of Law, Professor Judge, Cassation Administrative Court within the Supreme Court (Ukraine) Scientific Expert, CEPEJ-GT-CYBERJUST, Council of Europe ORCID: 0000-0002-2353-4836

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 Ukrainian Administrative Case Law as of 3 September 2026 English Edition, Version 1.0 DOI: https://doi.org/10.5281/zenodo.23205043 This is the author's English translation of the Ukrainian original, “Штучний інтелект та автоматизовані системи у публіч ному адмініструванні й судочинстві: людський контроль, мотивованість, перевірюваність та відповідальність у практиці Касаційного адміністративного суду у складі Верховного Суду ”, Version 1.0, Zenodo, 2026, DOI: https://doi.org/10.5281/zenodo.22274766. In the event of any discrepancy, the Ukrainian text shall prevail. Explanatory notes added specifically for this English edition are identified as such.

Abstract This analytical review examines the case law of the Cassation Administrative Court within the Supreme Court (Ukraine) concerning artificial intelligence, automated systems and legally significant outputs generated by digital systems in public administration and adjudication. Based on fourteen judicial acts delivered between 2023 and 2026, the review identifies several interconnected approaches: professional responsibility for the use of generative AI; responsibility of public authorities for automated administrative outcomes; allocation of functions between information systems and public officials; requirements concerning data reliability and verifiability; the evidentiary value of automatically generated

system records; and the limits of formalism in electronic proceedings. The analysis shows that automation may alter the technological means by which a professional or administrative function is performed, but it does not create an independent bearer of legal responsibility. Responsibility depends on the legal allocation of powers and on the source of the defect, including the rules or algorithms governing the system, the accuracy of primary data and the subsequent legal use of the generated output. The review also 1

highlights the role of human oversight, reason-giving, verifiability and judicial review in preventing accountability gaps in digitalised legal procedures.

Keywords: artificial intelligence; generative artificial intelligence; automated decision- making; automated public administration; human oversight; administrative justice; reason- giving; verifiability; digital evidence; accountability; legal responsibility; Supreme Court (Ukraine).

Note on the Ukrainian Legal Context and Terminology [Note to the English edition] The Supreme Court is the highest court in the Ukrainian judicial system. It comprises the Grand Chamber and four cassation courts, including the Cassation Administrative Court (CAC), which reviews administrative cases on points of law. Under Article 13(6) of the Law of Ukraine “On the Judiciary and the Status of Judges”, conclusions on the application of legal provisions set out in judgments (postanovy) of the Supreme Court must be taken into account by other courts.

In this Review, “judgment” renders postanova, a decision determining a cassation appeal, whereas “ruling” renders ukhvala, a procedural judicial decision. Rulings are not covered by Article 13(6). Case numbers (for example, 240/14153/24) identify the case across judicial instances; numbers beginning with K/990/ identify the cassation proceedings. All judicial acts cited in this Review are publicly available in Ukrainian in the Unified State Register of Court Decisions (reyestr.court.gov.ua).

INTRODUCTION

The use of artificial intelligence, automated information systems and software-based data-processing tools is gradually becoming a routine element of professional legal practice, public administration and adjudication. Digital systems verify and cross-check information, determine risk levels, generate documents, record procedural actions and provide electronic notification, while generative artificial intelligence is already being used to search for information and prepare legal texts.

Changes in technology do not, however, displace the requirements of legality, reliability, good faith, adequate reason-giving and judicial oversight. The question is not merely whether a particular technology may be used, but, above all, who is responsible for a legally significant output produced with its assistance; when that output requires human verification; what legal significance attaches to errors in an algorithm or in primary data; and what limits apply to intervention by a public official in an automated procedure defined by law.

These questions are of particular importance in administrative law. An automated system does not operate outside the competence of the authority that uses it, and the digital manner in which an administrative function is performed does not alter the constitutional 2

requirement, laid down in Article 19(2) of the Constitution of Ukraine, that a public authority act on the basis of, within the limits of, and in the manner prescribed by law.

Ukrainian legislation expressly reflects this logic. In particular, Article 62 of the Law of Ukraine “On Administrative Procedure”, No. 2073-IX of 17 February 2022, in force since 15 December 2023, permits administrative proceedings, in cases specified by law, to be conducted automatically, that is, by means of software without human intervention, while placing responsibility for administrative acts adopted in that mode on the administrative authority. Article 53(2) of the same Law recognises, among the means of evidence, data from national electronic information resources and the results of automated processing or verification of data.

In the case law of the Cassation Administrative Court within the Supreme Court (hereinafter, the CAC), these issues have developed along several interrelated lines: professional responsibility for the use of generative artificial intelligence; responsibility of a public authority for an automated administrative outcome; allocation of functions between an information system and a public official; requirements concerning the reliability and verifiability of data; the evidentiary value of automatically generated system records; and the limits of formalism in electronic proceedings.

This Review is based on an analysis of fourteen judicial acts of the Cassation Administrative Court within the Supreme Court delivered between 2023 and 2026. The judicial acts were selected according to their direct or contextual relevance to the use of artificial intelligence, automated systems, digital data, human oversight and responsibility.

The Review reflects the state of the case law as of 3 September 2026 and does not purport to cover exhaustively every decision of the CAC addressing digitalisation.

According to the degree of their direct connection with the subject matter of the Review, the judicial acts fall into two groups. The first comprises decisions in which the use of artificial intelligence, an automated system, automated data processing or a system- generated digital output directly formed part of the disputed legal relationship. The second comprises decisions used to identify general standards of reason-giving, data reliability, evidence, proper administration and judicial review that are also relevant to digitalised procedures. This distinction avoids equating the digital form of a legal relationship with the existence of a specific legal holding on automation.

The case law examined reveals a common approach: automation may alter the manner in which a professional or administrative function is performed, but it does not create a separate bearer of legal responsibility. Determining responsibility for an automated output requires identifying the nature of the defect and establishing who, under the law, is responsible for the rules governing the system, the reliability of the primary data, or the legal use made of the generated output.

3

1. Use of Artificial Intelligence in Judicial Proceedings: Verification of Outputs and User

Responsibility 1.1. Ruling of 15 January 2026 in Case No. 240/14153/24 Ruling of 15 January 2026 in Case No. 240/14153/24 refusing to open cassation proceedings, cassation proceedings No. K/990/56043/25 — https://reyestr.court.gov.ua/Review/133336040 This case directly concerned the use of generative artificial intelligence in preparing a procedural document. The materials submitted to the Supreme Court contained references to judicial decisions whose existence could not be confirmed. The Court linked this situation to the risk of using generative artificial intelligence without proper professional verification of the resulting output.

The Supreme Court did not proceed on the basis that the use of AI as an auxiliary tool is prohibited. The decisive point is different: its output must be verified by the person who uses it in professional activity. A generative tool does not replace legal analysis, source verification or the personal responsibility of the author of a procedural document for the information contained in it.

This approach also has a procedural dimension. Filing a document containing fabricated judicial decisions or other unverified information may go beyond an ordinary mistake and may be assessed in terms of good-faith exercise of procedural rights, proper performance of professional duties and the requirements of Article 45 of the Code of Administrative Procedure of Ukraine, which prohibits abuse of procedural rights. The Court also stated that filing procedural documents generated with the use of AI without proper professional verification may, in appropriate circumstances, indicate disrespect for the Court.

Accordingly, the legally significant issue is not the mere fact that generative artificial intelligence was used, but the absence of proper verification of its output and the subsequent use of unreliable information in a procedural document.

1.2. Ruling of 3 February 2026 in Case No. 340/2365/25 Ruling of 3 February 2026 in Case No. 340/2365/25 refusing to open cassation proceedings, cassation proceedings No. K/990/2939/26 — https://reyestr.court.gov.ua/Review/133764626 In this case, the Supreme Court established that the numbers and dates of judicial decisions cited in the cassation appeal had been fabricated and that the legal conclusions attributed to the Supreme Court were inaccurate. The Court expressly stated that these circumstances could indicate the use of automated text-generation tools, including artificial intelligence, without proper verification of the reliability of the resulting output.

4

Of particular importance is the rule formulated by the Court concerning the responsibility of the author of a document: irrespective of the method used to prepare it, responsibility for the accuracy of the information presented rests with the person who applies to the court. The technology used to prepare the text does not alter the bearer of procedural obligations and does not shift the risk of unreliable content onto the software.

The ruling develops the approach formulated in Case No. 240/14153/24. Taken together, these cases show that the use of generative AI in procedural activity is treated not as a separate form of legal activity, but as the use of an auxiliary tool whose output must undergo professional verification before it is relied upon in court.

International context [Note to the English edition]. Cases Nos. 240/14153/24 and 340/2365/25 have also been included in Damien Charlotin’s AI Hallucination Cases database, an international database tracking judicial decisions in which courts address fabricated or otherwise hallucinated content associated with generative artificial intelligence. At the time of preparation of this English edition, these were the two Ukrainian cases recorded in the database. Their inclusion provides a comparative reference point for situating the emerging Ukrainian case law within the broader transnational judicial response to fabricated AI-generated legal authorities in court proceedings. See: Damien Charlotin, AI Hallucination Cases Database,

https://www.damiencharlotin.com/hallucinations/ 1.3. Ruling of 19 June 2025 in Case No. 520/6119/23 Ruling of 19 June 2025 in Case No. 520/6119/23 refusing to open cassation proceedings, cassation proceedings No. K/990/17781/25 — https://reyestr.court.gov.ua/Review/128263149 An ancillary aspect of AI use arose in a case concerning an application to reinstate a missed procedural time limit. The party explained the delay, among other things, by the need to formulate a legal position, identify relevant case law and obtain access to a paid version of an artificial intelligence tool.

The Supreme Court did not regard those circumstances as an objective and insurmountable obstacle to taking the procedural step in time. Difficulties in choosing a legal position, searching case law, accessing databases or needing to use a particular AI tool concern the organisation of the party’s own activity and do not, in themselves, constitute a valid reason for missing a procedural time limit.

This approach complements the two preceding cases. A person is responsible not only for the accuracy of an output obtained with the assistance of AI, but also may not shift onto the chosen method of preparing a document the risks associated with untimely performance of that person’s own procedural obligation.

5

The case law of the CAC thus reveals a consistent approach: artificial intelligence may serve as a professional auxiliary tool, but it does not replace human verification and does not transfer to the technology responsibility for the content, reliability or timeliness of procedural activity.

2. Responsibility for an Algorithmically Generated Administrative Outcome

2.1. Judgment of 20 May 2026 in Case No. 280/9417/24 Judgment of the CAC of 20 May 2026, Case No. 280/9417/24, cassation proceedings No.

K/990/20592/25 — https://reyestr.court.gov.ua/Review/136682961 This judgment is central to understanding the CAC’s approach to automated public administration.

The dispute arose from information generated in the Unified State Electronic Database on Education (EDEBO) concerning the sequence in which education had been pursued. The automated output had direct legal significance for the individual, including in relation to entitlement to deferment from conscription during mobilisation.

In paragraphs 81–99 of the judgment, particularly paragraphs 88–93, the Supreme Court examined the relationship between digitalisation of administrative procedure and the responsibility of the competent authority. The use of automated tools does not release a public authority from the requirements of Article 19(2) of the Constitution of Ukraine, the duty to act lawfully, or the obligation to ensure the reliability of legally significant information.

Software-based data-processing tools are auxiliary in nature. An automated output must not replace legal assessment of the circumstances where such assessment is required for the correct application of the law. Digital technologies should facilitate the performance of an administrative function, not provide a basis for the competent authority to withdraw from responsibility for the consequences of that function.

Of particular significance is the conclusion that responsibility for an unlawful outcome cannot be shifted onto a technical algorithm or an educational institution that has no authority to alter its parameters where, under the applicable regulatory framework, the Ministry of Education and Science of Ukraine itself determines the requirements governing the operation of the system and is responsible for the exercise of the relevant powers.

For the purposes of this Review, the judgment is pivotal because it brings together, within a single legal approach, human oversight, data reliability, responsibility of the competent authority, the limits of automation and the requirement that a legally significant output comply with the law.

2.2. Judgment of 26 November 2025 in Case No. 160/29658/24 6

Judgment of the CAC of 26 November 2025, Case No. 160/29658/24, cassation proceedings No. K/990/28815/25 — https://reyestr.court.gov.ua/Review/132102704 In this case, the Supreme Court examined the allocation of competence among the Ministry of Education and Science of Ukraine, the technical administrator of EDEBO and educational institutions in connection with information generated by the system in an automated mode.

In paragraphs 58–62 of the judgment, the Court established that the relevant certificate is generated automatically in accordance with algorithms embedded in EDEBO. The Ministry of Education and Science of Ukraine determines the software requirements and the algorithms used to generate the information, while the technical administrator ensures the technical operation of the system and educational institutions are not authorised to alter its software parameters or algorithms.

In those circumstances, the duty to ensure that the disputed output is generated properly attaches to the entity to which the relevant powers are assigned by law. The Supreme Court identified the Ministry of Education and Science of Ukraine as that entity.

The judgment makes it possible to refine the broader proposition that, when determining responsibility for an automatically generated administrative outcome, it is necessary to establish who defines the legal and technical rules according to which the system generates legally significant information.

2.3. Judgment of 24 April 2025 in Case No. 280/3642/23 Judgment of the CAC of 24 April 2025, Case No. 280/3642/23, cassation proceedings No.

K/990/77/25 — https://reyestr.court.gov.ua/Review/126854551 The case concerned the imposition of administrative-economic sanctions for failure to comply with the statutory quota of jobs for persons with disabilities. The information required to determine whether grounds for the sanction existed was processed automatically on the basis of data from state information resources.

Paragraphs 81–85 of the judgment are particularly significant. Taking into account that the Law of Ukraine “On Administrative Procedure” entered into force after the disputed legal relationship had arisen, the Supreme Court referred to its provisions as a normative benchmark for automated administrative procedure. Article 53(2) of the Law permits data from national electronic information resources and the results of automated processing or verification of data to be used as evidence. Article 62 permits automated administrative proceedings and at the same time expressly places responsibility for administrative acts adopted in an automated mode on the administrative authority.

7

Automation therefore changes the technology through which an administrative function is performed, but does not, in itself, change the person or authority that the law designates as responsible for the administrative act.

At the same time, the case illustrates the other side of risk allocation. In paragraphs 75– 78, the Supreme Court drew attention to the employer’s responsibility for the accuracy of the primary reporting transmitted to state information resources. The Court also considered it necessary to establish whether the person had taken appropriate steps to correct the information submitted within the relevant time limit.

Accordingly, the verifiability of an automated output does not eliminate a person’s responsibility for the accuracy of primary data that the law requires that person to provide.

The judgment did not finally determine the dispute on the merits: the decisions of the lower courts were set aside and the case was remitted for fresh consideration. Its significance for this Review therefore lies primarily in the statutory provisions applied by the Court and in its approach to distinguishing responsibility for an automated act from responsibility for the primary data on which that act is based.

2.4. Judgment of 4 July 2024 in Case No. 520/16250/23 Judgment of the CAC of 4 July 2024, Case No. 520/16250/23, cassation proceedings No.

K/990/825/24 — https://reyestr.court.gov.ua/Review/120186679 The dispute concerned the automated generation of an extract from technical documentation on the normative monetary valuation of a land plot in which an incorrect coefficient had been applied. The legally significant output was generated using software and a geographic information system.

The Supreme Court proceeded on the basis that the use of software does not relieve the competent authority of its duty to ensure that the document generated complies with the applicable regulatory framework. Where an information system produces an output contrary to the prescribed rules, the authority cannot justify its lawfulness merely by relying on the fact that the calculation was performed automatically.

For the purposes of this Review, the judgment confirms a broader approach: an automated system is a means by which administrative activity is carried out, not an independent bearer of public powers. An error in a software calculation does not create an “accountability gap” between the individual and the authority using the system.

2.5. Allocation of Risk According to the Source of the Error Comparison of Cases Nos. 280/9417/24, 160/29658/24, 280/3642/23 and 520/16250/23 permits an important distinction to be drawn.

8

Where an incorrect outcome is caused by rules or an algorithm that, under the legal allocation of competence, are defined or controlled by the competent authority, the risk of that defect cannot be shifted onto the individual, the technical administrator or another entity that does not determine the relevant algorithm. Where, by contrast, the defect originates in primary information that the law requires the individual to provide, the relevant question is whether that person fulfilled the duty to ensure the accuracy of the data and to correct it in a timely manner.

Responsibility therefore depends not on the mere fact that the outcome was generated automatically, but on the source of the error and the legal allocation of duties among the participants in the relevant procedure.

3. Automated Administrative Procedure: Allocation of Functions Between the System and

the Public Official 3.1. Judgment of 24 October 2023 in Case No. 420/11131/22 Judgment of the CAC of 24 October 2023, Case No. 420/11131/22, cassation proceedings No. K/990/9882/23 — https://reyestr.court.gov.ua/Review/114524417 The dispute arose after the Department of State Architectural and Construction Control of Odesa City Council returned for revision a notice of 13 June 2022, No. OD051220613697, concerning commencement of construction works on a facility with minor consequences (consequence class CC1). The reasons given were the absence of the design documentation code and an inconsistency concerning the date of the expert report.

In paragraphs 47–59 of the judgment, the Supreme Court analysed the operation of the Unified State Electronic System in the Construction Sector (USESCS). The Procedure approved by Resolution of the Cabinet of Ministers of Ukraine No. 681 of 23 June 2021 “Certain Issues of Ensuring the Functioning of the Unified State Electronic System in the Construction Sector” (“Procedure No. 681”) provided for automated analysis of submitted data for completeness, format compliance, logic and reliability, determination of a risk level and, depending on that level, the type of procedure applicable to the document. A low risk level entailed automatic registration without the participation of a public official;

medium and high risk levels entailed consideration with the participation of public officials; and a critical risk level resulted in automatic blocking of the possibility of submitting the document.

The Court established that the claimants’ notice did not fall within the critical risk category and therefore Procedure No. 681 did not provide for it to be blocked; it was required to be registered in the Register of Construction Activities. The legislation likewise did not confer on an official of the state architectural and construction control authority the power to return such a notice for revision.

9

The significance of the judgment lies in the proposition that a legally prescribed operating procedure for an automated system not only supports the activity of an administrative authority but may also define the limits of permissible action by a public official. Human intervention cannot be used to create an additional procedure or form of response not provided for by law.

3.2. Judgment of 11 March 2025 in Case No. 460/22384/23 Judgment of the CAC of 11 March 2025, Case No. 460/22384/23, cassation proceedings No. K/990/44317/24 — https://reyestr.court.gov.ua/Review/125756659 The dispute concerned refusal to issue planning conditions and restrictions for the construction of an underground car park. One of the authority’s arguments was that the designated use of the land plot did not correspond to the parameters verified by USESCS and that, as a result, there was no technical possibility of producing a positive outcome for the administrative service.

In paragraphs 47–71 of the judgment, the Supreme Court examined the automated verification mechanism in detail. When an application is prepared and submitted, the system automatically analyses the data for completeness, format compliance, logic and reliability, determines the risk level and, where the level is critical, blocks submission of the document without registrar involvement.

In paragraph 68, the Court emphasised that the principal mechanism is automated verification by the system’s software. Registrar intervention is permitted where the system is unable to verify the specified criteria automatically. At the same time, the registrar is not authorised to alter a critical risk level determined automatically.

In the particular case, the claimant’s application had been accepted by the system and submitted through the Diia state e-services portal, which was relevant to assessing the authority’s assertion that the procedure could not technically be carried out. The Supreme Court also held that the conclusion that the designated use of the land plot had to be changed was erroneous, set aside the judgment of the appellate court and reinstated the first-instance judgment allowing the claim.

The judgment illustrates the safeguarding function of properly regulated automation. In cases defined by law, software-based verification not only accelerates data processing but also standardises the procedure and reduces the scope for arbitrary departure by a public official from established rules.

4. Reason-Giving, Data Reliability and Verifiability of a Legally Significant Output

For the purposes of this Review, verifiability means the ability to establish the factual data and legal grounds on which a legally significant output is based, assess the accuracy of 10

the information used, and identify the person or authority responsible for the decision taken or document generated. This approach does not create a separate requirement unknown to administrative law. In judicial practice, the relevant standard is expressed through established requirements of justification, reason-giving, reliability, completeness in establishing the circumstances, and the possibility of judicial review.

4.1. Judgment of 11 September 2023 in Case No. 420/14943/21 Judgment of the CAC of 11 September 2023, Case No. 420/14943/21, cassation proceedings No. K/990/8225/23 — https://reyestr.court.gov.ua/Review/113405942 This was not a case concerning automated decision-making. It concerned the lawfulness of a refusal by the state architectural and construction control authority to issue a certificate confirming that a facility was ready for operation. The electronic service was used only as a means of submitting documents.

For the purposes of this Review, the judgment is contextually relevant because of the standard formulated by the Supreme Court for the justification of decisions taken by public authorities. The Court linked adequate reason-giving, among other matters, to the logic and structure of the reasons, their connection with specific rules of law, legal assessment of the factual circumstances, consistency of the conclusions with the facts established, and the absence of conclusions unsupported by adequate reasons.

This standard is also relevant to digitalised administration. The use of an information system or automated data processing does not remove the authority’s duty to provide intelligible legal reasons for a decision where the law requires an individual assessment of the circumstances.

4.2. Judgment of 7 July 2026 in Case No. 460/14758/24 Judgment of the CAC of 7 July 2026, Case No. 460/14758/24, cassation proceedings No.

K/990/20442/26 — https://reyestr.court.gov.ua/Review/138037920 The dispute concerned four certificates issued on 1 August 2024 by the Rivne Regional Office of the Security Service of Ukraine (SSU) concerning the amount of remuneration for pension recalculation, in which the allowance for specific conditions of service was set at 52 per cent and the bonus at 10 per cent. The lower courts regarded those figures as lawful because they did not exceed the maximum limits established by law.

In paragraphs 62–71, the Supreme Court found that approach insufficient. The decisive issue was not only compliance with the prescribed maximum limits, but also determination of the actual average amounts of the relevant payments for the comparable position as of January 2020, 2021, 2022 and 2023. In paragraph 71, the Court stated that the relevant figures had to be determined on the basis of proper, reliable and verified data. In 11

paragraphs 81–89, the Court concluded that the lower courts had not fully established the circumstances of the case, set aside their decisions and remitted the case for fresh consideration.

For the purposes of the Review, this judgment is contextually relevant. A legally significant output does not become lawful merely because it formally falls within prescribed limits; it must be based on the factual data that the law identifies as relevant. This standard is particularly important for digitalised procedures in which an outcome depends on datasets and the accuracy of their processing.

Because the case was remitted for fresh consideration, this Review relies only on the requirements formulated by the cassation court concerning the reliability and verification of relevant data, without assessing the ultimate resolution of the dispute.

5. Digital Records, Electronic Evidence and Judicial Verification

5.1. Judgment of 26 August 2026 in Case No. 160/34163/25 Judgment of the CAC of 26 August 2026, Case No. 160/34163/25, cassation proceedings No. K/990/25589/26 — https://reyestr.court.gov.ua/Review/139254792 The dispute concerned determination of the time at which a judicial decision was served through the Unified Judicial Information and Communication System (UJICS) and the consequences of an automatically generated confirmation that it had been delivered to a personal electronic account (“e-cabinet”). A party disputed proper receipt of the document and relied on incorrect identification data.

The Supreme Court proceeded on the basis that an “e-cabinet” is a personalised means of electronic interaction and that legal significance therefore attaches to delivery of the document to the e-cabinet of the particular person on whom it was required to be served.

Where there are specific and substantiated objections concerning proper identification of the addressee, the mere fact that the system automatically generated a delivery confirmation is insufficient.

In paragraphs 56–73 and 85–102 of the judgment, the Court examined the significance of identification data and the need to verify information generated by the automated system.

Paragraph 100 contains the key conclusion for the case: when determining compliance with a procedural time limit, the decisive issue is not the mere fact that the system generated a confirmation, but whether it has been established that the document was in fact delivered to the e-cabinet of the proper participant in the proceedings.

An automatically generated system record may therefore serve as evidence of a legally significant event; however, where specific circumstances cast doubt on the accuracy of its generation or the identification of the addressee, the record is subject to substantive 12

verification. Automation of evidentiary processes does not make the automated output irrefutable. The case was remitted to the appellate court for continued consideration. Accordingly, this Review relies only on the procedural standard formulated by the Supreme Court for verifying proper electronic service, rather than on any assessment of the merits of the dispute.

5.2. Judgment of 20 June 2023 in Case No. 420/4540/22 Judgment of the CAC of 20 June 2023, Case No. 420/4540/22, cassation proceedings No.

K/990/32869/22 — https://reyestr.court.gov.ua/Review/111724996 In this case, the Supreme Court assessed digital information, including data from electronic mapping services, in the context of proving factual circumstances.

Its significance for this Review lies not in the formulation of a special standard for automated administrative decision-making, but in recognition that digital data and information created or displayed by electronic services form part of the evidentiary environment of contemporary administrative adjudication and must be assessed together with other evidence under the general rules of the Code of Administrative Procedure of Ukraine.

The digital origin of information does not, in itself, determine its reliability or sufficiency. What remains decisive is the court’s assessment of the relevance, admissibility and reliability of the evidence and of its relationship with the other circumstances established in the case.

5.3. Judgment of 1 May 2023 in Case No. 540/913/21 Judgment of the CAC of 1 May 2023, Case No. 540/913/21, cassation proceedings No.

K/9901/40933/21 — https://reyestr.court.gov.ua/Review/110568942 This judgment is used in the Review as a contextual source concerning the legal fiction of notification and the inadmissibility of excessive formalism.

The dispute arose from a sanction imposed on a business entity by the National Energy and Utilities Regulatory Commission (NEURC) and concerned, among other matters, whether the entity had been properly notified. In paragraphs 75–81, the Supreme Court distinguished between actual notification and a legal fiction of notification to which the law attaches specified legal consequences. The Court also emphasised the inadmissibility of excessive formalism and “legal purism” when procedural violations are assessed.

For digital procedures, this approach is of supporting relevance. Electronic or automated recording of a legally significant fact may produce the consequences prescribed by law only where the statutory conditions attached to the relevant legal fiction are satisfied. At 13

the same time, this judgment does not itself establish a special standard governing the evidentiary weight of an automated system record; that standard is reflected much more directly in Case No. 160/34163/25.

6. Synthesis of the Approaches Developed in the Case Law

Analysis of the case law of the CAC makes it possible to distinguish at least three levels in the allocation of legally significant duties and responsibility connected with the use of artificial intelligence and automated systems.

The first level concerns the user of the technology. A person who uses generative artificial intelligence to prepare a procedural document is responsible for verifying the output, for the accuracy of the information presented, for good-faith procedural conduct and for compliance with time limits.

The second level concerns the public authority that uses an automated system to exercise powers defined by law. Automatic generation of a document or indicator does not release the authority from the duties imposed on it by law where that authority itself defines or controls the rules governing the operation of the system or is responsible for the legal use of the system-generated output in the exercise of public powers.

The third level concerns the source of the primary data. Where the law places on a person the duty to provide accurate information, automated processing of that information by the state does not remove that person’s responsibility for its accuracy. Identifying the party that bears the risk of error therefore requires establishing not only who uses the automated system, but also where the defect arose: in the rules governing the system, in the primary data, or in the subsequent legal use of the generated output.

The case law concerning USESCS further demonstrates that automation may have not only risk-related but also safeguarding effects. Where legislation clearly specifies which criteria are checked automatically, what risk level they produce and when intervention by a public official is permitted, automation limits the scope for arbitrary alteration of the procedure or creation of additional grounds for an adverse decision.

At the same time, an automatically generated digital record does not acquire absolute evidentiary force merely by virtue of its origin. Where a person identifies specific circumstances that cast doubt on the accuracy of an automated output, the court must assess those circumstances and, where necessary, verify the information held in the system.

Automation does not create an independent bearer of legal responsibility. Determining responsibility for an automated output requires identifying the person or authority that, under the law, is responsible for the rules governing the system, the reliability of the primary data, or the legal use made of the generated output.

14

Comparative significance [Note to the English edition]. The significance of this case law extends beyond the particular Ukrainian systems examined. The cases illustrate a recurring problem of digital public administration: how legal responsibility should be allocated where a legally significant result is produced through a combination of statutory rules, software architecture, primary data and subsequent human use of a system-generated output. The approach emerging from the Ukrainian case law places responsibility not on the technology itself, but on the legally defined allocation of competence and on the source of the relevant defect.

CONCLUSIONS

1. Automation changes the manner in which a public or professional function is performed,

but it does not remove duties and responsibility imposed by law on the competent person or authority. Mere reliance on software, an algorithm or generative artificial intelligence does not release a person or authority from legal duties concerning the reliability, lawfulness and adequate reason-giving of a legally significant output.

2. The use of generative artificial intelligence in legal practice is permissible as an auxiliary

tool, provided that its output is subject to proper professional verification. The person who files a procedural document with the court remains responsible for its content and for the accuracy of the case law and other information cited in it. Use of AI without verification may have consequences for the good-faith character of procedural conduct and the proper performance of professional duties.

3. In automated public administration, responsibility is determined by the legal allocation

of competence, not by the technical manner in which an output is generated. Where an incorrect output results from rules or an algorithm that the competent authority defines or controls under the applicable legal allocation of competence, the risk of that defect cannot be shifted onto the individual, the technical administrator or another entity that lacks the relevant power to define those rules or that algorithm.

4. Allocation of the risk of error depends on the source of the defect. An error in rules or

an algorithm for which a public authority is responsible under the legal allocation of competence should not be shifted onto an individual or technical administrator that does not define those rules. Where an incorrect output results from unreliable primary data that a private party is legally required to provide, the relevant consideration is whether that party properly fulfilled the duty to ensure the accuracy of the data and to correct it in a timely manner.

5. Automation may perform a safeguarding function. Legally prescribed automated

verification can standardise administrative procedure, promote consistent application of criteria and limit arbitrary intervention by a public official. Human intervention remains 15

necessary, however, where the law requires an individual assessment or where the system itself is unable to verify legally significant circumstances.

6. An automatically generated output or system record is not irrefutable merely because it

is digital in origin. Where there are specific and substantiated doubts concerning the accuracy of the data, identification of the person or compliance with the applicable legal conditions, the output must be verified under the general rules of administrative adjudication.

7. Digitalisation does not reduce the requirements of reason-giving and reliability

applicable to an administrative decision. A legally significant output must be based on proper and verified data and, where the law provides for an individual decision by the authority, must contain sufficient reasons to enable the person concerned to understand the grounds for the decision and the court to review its lawfulness.

8. Automated public administration and electronic justice should develop without creating

“accountability gaps” and without excessive procedural formalism. The technological means by which a function is performed should strengthen, rather than weaken, legality, data reliability, the possibility of correcting errors and effective judicial review.

List of Judicial Acts Analysed All judicial acts listed below were delivered by the Cassation Administrative Court within the Supreme Court and are available in Ukrainian in the Unified State Register of Court Decisions.

1. Ruling of 15 January 2026 in Case No. 240/14153/24, cassation proceedings No.

K/990/56043/25 [in Ukrainian] — https://reyestr.court.gov.ua/Review/133336040

2. Ruling of 3 February 2026 in Case No. 340/2365/25, cassation proceedings No.

K/990/2939/26 [in Ukrainian] — https://reyestr.court.gov.ua/Review/133764626

3. Ruling of 19 June 2025 in Case No. 520/6119/23, cassation proceedings No.

K/990/17781/25 [in Ukrainian] — https://reyestr.court.gov.ua/Review/128263149

4. Judgment of 20 May 2026 in Case No. 280/9417/24, cassation proceedings No.

K/990/20592/25 [in Ukrainian] — https://reyestr.court.gov.ua/Review/136682961

5. Judgment of 26 November 2025 in Case No. 160/29658/24, cassation proceedings

No. K/990/28815/25 [in Ukrainian] — https://reyestr.court.gov.ua/Review/132102704

6. Judgment of 24 April 2025 in Case No. 280/3642/23, cassation proceedings No.

K/990/77/25 [in Ukrainian] — https://reyestr.court.gov.ua/Review/126854551

7. Judgment of 4 July 2024 in Case No. 520/16250/23, cassation proceedings No.

K/990/825/24 [in Ukrainian] — https://reyestr.court.gov.ua/Review/120186679 16

8. Judgment of 24 October 2023 in Case No. 420/11131/22, cassation proceedings

No. K/990/9882/23 [in Ukrainian] — https://reyestr.court.gov.ua/Review/114524417

9. Judgment of 11 March 2025 in Case No. 460/22384/23, cassation proceedings No.

K/990/44317/24 [in Ukrainian] — https://reyestr.court.gov.ua/Review/125756659

10. Judgment of 11 September 2023 in Case No. 420/14943/21, cassation proceedings

No. K/990/8225/23 [in Ukrainian] — https://reyestr.court.gov.ua/Review/113405942

11. Judgment of 7 July 2026 in Case No. 460/14758/24, cassation proceedings No.

K/990/20442/26 [in Ukrainian] — https://reyestr.court.gov.ua/Review/138037920

12. Judgment of 26 August 2026 in Case No. 160/34163/25, cassation proceedings No.

K/990/25589/26 [in Ukrainian] — https://reyestr.court.gov.ua/Review/139254792

13. Judgment of 20 June 2023 in Case No. 420/4540/22, cassation proceedings No.

K/990/32869/22 [in Ukrainian] — https://reyestr.court.gov.ua/Review/111724996

14. Judgment of 1 May 2023 in Case No. 540/913/21, cassation proceedings No.

K/9901/40933/21 [in Ukrainian] — https://reyestr.court.gov.ua/Review/110568942 Related Publications and Materials by the Author: All online sources last accessed: 7 October 2026.

1. Bernaziuk, Ian. Artificial Intelligence in the Ukrainian Judiciary: Charting the Course

Under the Digital Gavel. Constitutionalist.com.ua. 2025. URL: https://constitutionalist.com.ua/artificial-intelligence-in-the-ukrainian-judiciary-charting- the-course-under-the-digital-gavel

2. Bernaziuk, Ian. The AI Era and the Role of Supreme Courts in the Digital Transformation

of Justice [in Ukrainian]. Yurydychna Hazeta [Legal Gazette]. 2025. No. 4 (792). P. 16–18. URL: https://yur-gazeta.com/publications/practice/sudova-praktika/era-shi-y-rol- verhovnih-sudiv-u-cifroviy-transformaciyi-pravosuddya.html

3. Bernaziuk, Ian. The Principle of Professional Human Oversight as a Basis for Interaction

Between Justice and Artificial Intelligence [in Ukrainian]: presentation delivered at a training course for judges of local and appellate courts (Odesa, 15 October 2025) / Supreme Court; National School of Judges of Ukraine. URL: https://court.gov.ua/storage/portal/supreme/prezentacii_2025/147_Professional_Huma n_Oversight_Justice_AI_bernaziuk.pdf

4. Bernaziuk, Ian. AI in National Judicial Proceedings: Evolution of Legal Positions [in

Ukrainian]: presentation delivered at the scientific and practical training course “Use of Artificial Intelligence in Judicial Activity” (Kyiv, 5 March 2026) / Supreme Court; National School of Judges of Ukraine; Council of Europe. URL: 17

https://court.gov.ua/storage/portal/supreme/prezent2026/175_AI_in_National_Judicial_ Proceedings_Evolution_Legal_Positions_bernaziuk.pdf

5. Bernaziuk, Ian. AI and Automated Adjudication: National and International Standards [in

Ukrainian]: presentation delivered at the IX International Legal Forum “Human Rights and Public Governance” (Chernivtsi, 29 May 2026) / Supreme Court. URL: https://court.gov.ua/storage/portal/supreme/prezent2026/193_AI_Automated_Adjudicat ion_NationalInternational_Standards_bernaziuk.pdf

6. 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

7. 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: https://doi.org/10.5281/zenodo.23103501 Transparency Declaration on the Use of Artificial Intelligence In preparing the Ukrainian original of this Review, artificial intelligence tools were used as an auxiliary means for structural and linguistic editing of the text, checking its internal logic, identifying repetition, and considering alternative formulations of individual passages.

Artificial intelligence was not used as an independent source of legal conclusions and did not replace the author’s determination of the subject of the study or analysis of the judicial decisions. Information concerning the content, details and legal conclusions of the judicial acts was verified against the full texts of the respective decisions.

The concept of the Review, the selection and legal assessment of the case law, determination of its significance, formulation of the author’s syntheses and conclusions, and the final editorial decisions belong to the author. Outputs obtained with the assistance of artificial intelligence were critically reviewed and were accepted, modified or rejected by the author.

The English edition was prepared by the author with the assistance of ChatGPT (OpenAI) and Claude (Anthropic), used as auxiliary tools for draft translation, linguistic editing and terminological consistency checks. All AI-assisted output was critically reviewed, revised and finalised by the author, who bears sole responsibility for the accuracy and final wording of the English text.

18

Status of the Review This Review is an independent scholarly and analytical work and does not constitute an official review or an official position of the Supreme Court, the Council of Europe or CEPEJ.

It is based exclusively on published judicial acts and other publicly available materials and does not rely on judicial deliberations, non-public court materials or information obtained by the author in a judicial capacity.

[Note to the English edition] The author served as judge-rapporteur in all fourteen cases analysed. The Review draws on judicial acts from the author's own docket that are directly or contextually relevant to its subject matter; it therefore does not purport to represent the entire body of CAC case law on these issues.

The syntheses presented here are the author’s scholarly and analytical interpretation of published judicial acts and do not constitute their official interpretation.

Suggested Citation: 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. English Edition, Version 1.0. Zenodo, 2026. DOI: https://doi.org/10.5281/zenodo.23205043 © 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/

19