Generative artificial intelligence is already used frequently by many students, but frequent use does not ensure verification, disclosure, privacy awareness, or protection against dependence. This study examined responsible-use practices, student-reported barriers, and institutional priorities among a newly enrolled bachelor-level cohort in Kathmandu, Nepal. A cross-sectional online survey received 74 consenting submissions; six uniform response patterns across 20 scale items were excluded, leaving 68 records for descriptive analysis. Although 62 respondents (91.2%) used generative AI daily or several times per week, only 27 (39.7%) always or often verified important generated information and 21 (30.9%) always or often disclosed AI use when disclosure was required. Privacy or data-security concerns were the most frequently selected barrier (39; 57.4%), followed by overdependence (35; 51.5%), inaccurate information (32; 47.1%), and the cost of paid tools (29; 42.6%). Students most often prioritized AI training for students and teachers (41; 60.3%), privacy and data-security guidance (38; 55.9%), clear generative AI policies (37; 54.4%), and improved internet and technical support (35; 51.5%). Of 33 substantive open comments, 18 requested practical AI-literacy training and 12 requested clearer rules or responsible-use guidance. The findings show a gap between high adoption and consistent critical practice. Institutions should combine first-year AI literacy, assessment-specific rules, standard disclosure processes, privacy safeguards, verification routines, technical support, and assessment design that preserves student reasoning. The evidence is descriptive and specific to one college cohort; it should not be generalized to all students in Kathmandu Valley or Nepal.