Objectives:
The objective of this study is to determine whether the use of artificial intelligence-based tools - specifically language models - can serve as a technically feasible and legally permissible form of support for the adjudicative activities of administrative courts, particularly regarding the verification of grounds for the rejection of a complaint as set forth in Article 58 of the Law on Proceedings before Administrative Courts (p.p.s.a.). The analysis aims to define the boundaries of permissible automation for analytical tasks within administrative court proceedings, while distinguishing between supporting the decision-making process and replacing the judge. The study also includes an empirical assessment of the potential for the algorithmic classification of legal criteria, involving the design of an application that analyzes the content of complaints and generates draft court orders. Furthermore, the research identifies the limitations of language models and evaluates the proposed solution in light of the requirements set out in Regulation (EU) 2024/1689.

Methods:
The study employs a doctrinal legal method, systemic and functional analysis, and an empirical approach. The research encompassed an analysis of the provisions of the Law on Proceedings before Administrative Courts (p.p.s.a.) and Regulation (EU) 2024/1689, as well as the design and testing of an application utilizing a language model. Furthermore, machine learning methods were applied, alongside a comparative analysis of the effectiveness of models designed to classify the grounds for the rejection of a complaint.

Results:
The study demonstrated that using language models to analyze complaints under Article 58 of the Law on Proceedings before Administrative Courts (p.p.s.a.) is technically feasible, albeit to a limited extent. These limitations stem primarily from the incompleteness of the textual data. While an AI system can support an initial formal-legal analysis, it cannot replace the process of legal subsumption performed by a judge and requires human oversight, in accordance with the EU AI Act.

Conclusions:
The study results indicate that algorithmic analysis of the content of complaints can, in some cases, identify grounds for rejection as set forth in Article 58 of the Law on Proceedings before Administrative Courts (p.p.s.a.). At the same time, significant informational limitations - stemming from a lack of procedural data - have been revealed. The findings confirm the utility of language models as a tool to support the preliminary formal-legal analysis of complaints in administrative court proceedings, particularly in identifying grounds for rejection that are discernible within the linguistic layer of a procedural document.