AI is moving from experimentation to scrutiny inside academic and institutional governance, with new guidance and oversight expectations emerging for both research and administrative decision-making. In the UK, panellists working on the next Research Excellence Framework (REF) 2029 are debating how to audit narrative research-environment claims—concerns that “polishing” could turn evidence standards into a compliance problem. Separately, commentary in scholarly publishing highlights a growing mismatch between the need to manage generative AI usage and the lack of reliable tools and procedures for detecting or properly evaluating AI-assisted work. The discussion centers on transparency challenges—what AI was used for matters, but broad disclosures often do not capture the intellectual contribution. Together, the REF scrutiny and the publishing ethics debate show institutions are racing to adapt oversight mechanisms for a world where AI can generate persuasive language, structured outputs, and partial analysis without clear accountability.