Research integrity education may demonstrate that these programs establish critical foundations for maintaining healthy academic ecosystems. The traditional educational approaches might indicate limitations including insufficient interactivity and low personalization. Moreover, this study could examine a generative conversational agent system for research integrity education. The system appears to enhance integrity literacy through dual behavioral intervention mechanisms. The research employed a quasi-experimental design with 126 graduate students. Given that participants received random assignment into experimental and control groups, the experimental group participated in an eight-week agent-based intervention. The control group received traditional lecture-based training instead. Furthermore, we measured indicators at baseline, immediate post-intervention, and three-month follow-up points. These indicators may show cognition level, behavioral norms, moral reasoning ability, and critical thinking. The results could demonstrate that the significant experimental group showed greater improvements than the control group across all important indicators. Research integrity cognition might indicate increases by 25.5%. Notwithstanding these critical empirical findings, the important academic behavioral norms could be improved by 25.1% substantially. Nonetheless, several limitations warrant acknowledgment: the sample was drawn from a single institution, limiting generalizability; the three-month follow-up period may be insufficient to capture long-term behavioral transfer; and the system's performance in handling culturally nuanced or highly ambiguous ethical dilemmas remains suboptimal. These findings contribute empirical evidence for the application of generative artificial intelligence in professional ethics education while offering practical guidance for the intelligent transformation of research integrity training in academic institutions.