Abstract: The subject of the research is the legal expertise of administrative regulation projects prepared using generative artificial intelligence (AI). The article discusses how to verify the compliance of applicant requirements with current legislation and the powers of the authority specifying the procedure for providing a service. The verification includes deadlines, lists of documents, grounds for refusal, and the procedure for appeals. A reliable reference to an active act does not yet confirm the legality of the proposed requirement. The action of norms over time and the necessity of preserving the materials for project preparation are taken into account, allowing for the verification of the model's proposals based on the sources provided to it. The aim of the article is to determine the features of legal expertise for administrative regulation projects prepared using generative AI and to justify the procedure for checking the legal foundations of the requirements contained therein. A formal-legal analysis of legislation and legal positions of courts on the legality of administrative regulations, legal certainty, and limits of subordinate norm-setting has been employed. Scientific works on the use of AI have been reviewed in relation to the tasks of project expertise. The competence of the authority, the conditions for the application of norms, and their action over time have been considered. A procedure for the legal expertise of administrative regulation projects prepared using generative AI has been proposed. It has been shown that the method of preparing the project does not change the criteria for its legality. A reliable reference to an active act does not yet confirm the legality of the requirement to the applicant: it is necessary to establish the content of the norm, the conditions for its application, its action over time, and the authority's powers. The regulation must maintain established exceptions by the norms and features of service provision to different categories of applicants. It is proposed to establish the developer's obligation to present for expertise the tasks for the AI, the sources provided to the model indicating revisions, and the justification for the acceptance or rejection of the model's proposals. Time savings should be assessed considering the costs of verifying erroneous recommendations and the risk of missing significant deficiencies.

Keywords: generative artificial intelligence, administrative regulations, legal review, agency rulemaking, public services, powers of public authorities, legal certainty, anti-corruption review, case law, applicant rights