Answers are grounded in this paper and cite their sources.
Session discusses AI's capabilities in academic settings, highlighting implications for integrity and assessment design.
Session overview: Universities increasingly distinguish between “acceptable” and “unacceptable” uses of artificial intelligence (AI), often permitting grammar and spelling correction while prohibiting content generation. However, such distinctions assume that generative AI systems can reliably restrict themselves to these limited functions. This session shows that this assumption might not be justified. The session begins by considering common sources of confusion surrounding institutional AI policies, drawing on examples of policy language that attempt to distinguish between proofreading, editing, and content generation. It then presents a series of experimental demonstrations, which I conducted, in which I explicitly instructed AI (GPT-5.5) to perform grammar and spelling only corrections on short psychology texts that I created. In these, I included both linguistic and factual errors. These examples were designed to mirror situations in which students seek permitted forms of AI assistance. The demonstrations reveal a range of unexpected AI behaviours, including correcting factual inaccuracies, reversing scientific claims, replacing theorists, introducing new references, applying disciplinary conventions, and providing unsolicited academic feedback. In some cases, the AI effectively acted as a tutor, editor, or co-author despite being explicitly asked only to proofread. I will demonstrate my experimental findings in the session. The session thus argues that these findings raise important questions about the enforceability, fairness, and coherence of current AI policies. The session concludes by discussing alternative approaches that focus on assessment design and demonstrated learning outcomes rather than attempts to regulate increasingly blurred categories of AI assistance. Key learning points from this session: Attendees will gain practical insight into how generative AI behaves when asked to provide “grammar-only” support and will consider the implications for academic integrity, fairness, and assessment design. The session will encourage critical reflection on whether current AI policies are enforceable, coherent, and sustainable.
No takes yet. Share an insight, caveat, or question.
Sylvia Terbeck (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: