Purpose This paper aims to examine the scope of legal responsibility arising from violations of the obligation to maintain the confidentiality of public office information, particularly when such violations occur through the use of generative artificial intelligence (AI) systems, with ChatGPT serving as a representative example. It seeks to evaluate the extent to which established criminal and disciplinary liability frameworks can effectively address technology-mediated disclosures and to assess whether these traditional mechanisms remain sufficient within the evolving landscape of digital governance. Design/methodology/approach The study adopts a doctrinal legal methodology combining analytical, descriptive, critical and applied approaches. It relies on an in-depth analysis of criminal and administrative legal rules governing public office confidentiality, alongside an examination of contemporary international academic literature on AI, data governance and public sector accountability. Findings The paper finds that breaches of confidentiality no longer occur solely through direct or intentional disclosure but increasingly through indirect, technical and probabilistic forms of exposure linked to data input and reliance on AI-generated outputs. It further demonstrates that while criminal liability faces significant doctrinal and evidentiary constraints, disciplinary liability offers a more flexible and effective framework for addressing such breaches in practice. Practical implications The findings underline the necessity for public authorities to adopt clear internal policies governing the use of ChatGPT and to prioritise training and awareness among public officials to mitigate confidentiality risks. Originality/value The paper contributes to the literature by bridging the gap between traditional doctrines of public office secrecy and the legal challenges posed by ChatGPT, offering a structured legal analysis that integrates criminal, disciplinary and governance perspectives.
Aboelazm et al. (Thu,) studied this question.