Governing generative AI in final academic works: from disclosure to competence verification in European university policies

Generative artificial intelligence has unsettled a central assumption in assessing final academic works: that the submitted text is reliable evidence of a student’s competence, authorship and intellectual contribution. European universities have responded with acceptable-use rules, disclosure requirements and integrity guidance, but rarely explain what evidence would show competence once research, writing and revision are partly machine-mediated. This article asks how far official policy supplies that evidence. Drawing on argument-based validity theory and work on evaluative judgement, it defines inferential governance and builds an Inferential Governance Framework of 15 indicators across five dimensions (competence, process, evidence, interpretive assessment and governance). It is applied to 70 official university policy cases from 14 European countries. Policy development proves substantial but uneven: 50 cases are procedural or declarative and two reach the highest of five bands. Declaration of AI use, competence-oriented language and independent intellectual contribution are widespread; process supervision, student-supervisor agreements, process dossiers, oral defence, the examiner or tribunal role and detector prudence are much less common. The article calls this imbalance an inferential gap between competence claims and the means of verification, shows that it survives every alternative corpus composition tested, and proposes a Competence-Centred Inferential Governance Model.