When ChatGPT Enters the Classroom: Student Use, Staff Perceptions, and Integrity Responses in Romanian Higher Education

Synopsis

Survey evidence from four Romanian universities (N=786 students; N=472 teaching staff) shows that disclosure norms for Gen-AI remain weak, with widespread uncertainty about whether/how to acknowledge AI use. Framed through an integrity-infrastructure lens, the study contrasts staff’s strong plagiarism classification with mixed student views and identifies integrity training/support as a protective factor.

Key questions

1. What disclosure patterns and uncertainties characterize student Gen-AI use?

2. How do students and staff define undisclosed Gen-AI in relation to plagiarism?

3. Which institutional supports and risk factors predict problematic AI use?

Keywords

Generative AI, academic integrity, higher education, neo-institutional theory, integrity infrastructure, policy clarity

The rapid diffusion of generative artificial intelligence (Gen-AI) tools such as ChatGPT is reshaping learning and assessment in higher education and, simultaneously, testing the resilience of academic integrity governance. While many universities have issued formal rules for AI use, growing evidence suggests a gap between policy design and everyday pedagogical practice (Sullivan et al., 2023). Building on neo-institutional theory, this study examines Gen-AI integration in Romanian higher education as a legitimacy-driven response that can generate institutional decoupling and a “normative vacuum” when rules remain ambiguous or weakly implemented.

We analyze data from a large-scale quantitative survey administered in 2025 at four major Romanian universities (N=786 students; N=472 teaching staff). Using descriptive analysis and ordinal logistic regression (PLUM), we map AI use across low- and high-stakes assessments and test associations between perceived policy clarity, integrity training/support, and self-reported AI-assisted misconduct. Results show extensive student use of AI for coursework (56.2% report using AI at least once for a paper/project) but much lower self-reported use in exams (83.8% report never using AI for test/exam answers). Disclosure remains inconsistent (only a minority clearly acknowledge AI use), and students often frame non-disclosed use as context-dependent rather than categorically plagiarism. Teaching staff, however, show strong normative consensus: 89.2% classify unauthorized, undisclosed student AI use as plagiarism. Across models, stronger integrity training/support and clearer policy communication are associated with lower problematic AI use, suggesting that integrity “infrastructure” matters more than purely punitive approaches.

Among students who used AI in a thesis/paper, only 15.3% report explicitly acknowledging AI use (7.9% partial acknowledgement), while 49.5% are unsure whether/how to acknowledge and 27.3% do not mention AI use at all. Perceptions of whether undisclosed AI use constitutes plagiarism are similarly fragmented: 35.5% answer “yes”, 9.0% “no”, and 55.5% report “sometimes/it depends,” indicating a situational ethics frame rather than a stable norm of attribution. For a high-risk practice, automatically generating bibliographies using AI without checking sources, most students report non-engagement (90.6% never), but a non-zero minority admits at least one occurrence (5.7% once; 2.6% several times).

In multivariate models, integrity education and support function as protective factors, with a composite integrity training/support measure showing a strong negative association with problematic AI use (β = −7.148, p < .05). By contrast, AI-assisted cheating in exams clusters in a high-risk subgroup already engaged in traditional dishonesty, reflected by very large positive coefficients for deviance indicators (e.g., β = 10.581; β = 8.262; p < .001).

Staff prevention strategies combine control and pedagogy: 31.4% assess “exclusively in class,” 28.1% use AI-capable anti-plagiarism tools, 14.0% discuss restrictions/rules, and 13.4% ask students to report AI tools used (10.7% report no specific measures). Drawing on an integrity infrastructure perspective that differentiates compliance controls from integrity-oriented training and support (Bertram Gallant, 2017; Bretag, 2016), we argue that institutionalizing clear disclosure guidance and assessment redesign is central to consolidating transparency.

References:

· Bertram Gallant, T. (2017). Academic integrity as a teaching & learning issue: From theory to practice. Theory Into Practice, 56(2), 88–94. https://doi.org/10.1080/00405841.2017.1308173

· Bretag, T. (2016). Defining academic integrity: International perspectives – Introduction. In T. Bretag (Ed.), Handbook of Academic Integrity (pp. 3–5). Springer. https://doi.org/10.1007/978-981-287-098-8_76

· Sullivan, M., Kelly, A., & McLaughlan, P. (2023). ChatGPT in higher education: Considerations for academic integrity and student learning. Journal of Applied Learning and Teaching, 6(1), 31–40. https://doi.org/10.37074/jalt.2023.6.1.17