AI Risk Awareness and Reported Rule Following among Law Students in Bangladesh

Abstract Understanding the risks of generative artificial intelligence (AI) and believing that regulatory awareness helps responsible use are distinct from reporting adherence to applicable rules. This cross-sectional study examined those distinctions among 133 respondents recorded as law students in a 385-record survey on AI governance in Bangladesh. We report full item distributions, joint endorsement and paired contrasts, and fit exploratory models of the original rule-following rating. Among law students, 108 (81.2%) endorsed avoiding unprotected disclosure of confidential or sensitive information, 69 (51.9%) endorsed verification before reliance or sharing and 64 (48.1%) endorsed following applicable institutional or professional rules. Confidentiality endorsement exceeded rule following by 33.1 percentage points (pointwise 95% CI 21.8 to 44.4; Holm-adjusted p < 0.001), while the verification difference did not meet the adjusted threshold. Of the 103 respondents endorsing the benefit of regulatory awareness, 51 also endorsed rule following. Self-reported understanding of ethical and legal risks, general AI awareness and perceived benefit of regulatory awareness did not meet the adjusted threshold in exploratory regression. The conclusion was unchanged in binary and full-sample sensitivity analyses. The findings distinguish a specific information-handling precaution from reported adherence to unspecified rules. They support evaluating whether students can locate, interpret and apply relevant rules in concrete tasks. The study does not measure demonstrated AI literacy, knowledge of regulations, audited conduct or the effect of education on compliance.