Beyond policy exposure: ethical-integrity appraisal and responsible generative AI intentions among students from two Malaysian universities

Introduction Universities increasingly communicate rules for responsible generative artificial intelligence (GenAI) use, but nominal exposure to guidance may not translate into students' intentions. Evidence is also weakened when policy exposure, ethical beliefs, and Theory of Planned Behavior (TPB) constructs are treated as interchangeable. This study examined whether students' appraisal of ethical responsibilities and academic-integrity risks was associated with responsible GenAI intention, and whether recalled lecturer discussion contributed additional explanatory value. Methods A cross-sectional online questionnaire was completed by 441 students recruited through convenience and snowball procedures from one public and one private Malaysian university. Seventy-one submissions met a fixed invariant-response rule, leaving 370 respondents in the primary analytic pool. Researcher-developed focal measures comprised six ethical-awareness items, seven perceived academic-integrity-risk items, and four responsible-intention items. Reliability, preliminary internal structure, ordinal robustness, group differences, HC3-robust regression, diagnostics, and full-sample sensitivity were evaluated. Results A correlated three-factor model separating ethical awareness, academic-integrity risk, and responsible intention fitted better than one-factor and two-factor alternatives. The two appraisal facets were strongly related, supporting their use as a transparently defined umbrella score while indicating limited discriminant separation. Ethical-integrity appraisal was moderately associated with responsible intention, r = 0.469, 95% CI [0.384, 0.547], and explained 22.0% of its variance, B = 0.555, HC3 SE = 0.074, p < 0.001. Intention did not differ significantly across lecturer-discussion responses, F (2, 361) = 2.18, p = 0.115, eta squared = 0.012. Discussion exposure added Delta R 2 = 0.003 beyond appraisal, HC3 F (2, 350) = 0.62, p = 0.538. Invariant responses inflated the full-sample association to r = 0.610. Discussion Ethical-integrity appraisal was associated with responsible-use intention, whereas a single indicator of lecturer-discussion exposure was not. The results support a bounded TPB-informed interpretation, not a full test of TPB or evidence of causal effects, internalised policy, or responsible behaviour. The study's main contribution is methodological and empirical: it distinguishes appraisal, communication exposure, and intention while showing how response-quality screening changes effect magnitude.