Abstract
<title>Abstract</title> <p>Background Generative artificial intelligence (GenAI) tools such as ChatGPT and DeepSeek are increasingly used in medical education, yet how medical graduate students perceive the risks, benefits, and responsible use of these tools in academic research remains underexplored, particularly in non‑Western contexts. Methods This descriptive qualitative study conducted face‑to‑face, semi‑structured interviews with 18 first‑year master’s students from five clinical specialties at a medical university in Shandong Province, China. Participants had all used GenAI for academic tasks at least once. Interviews were audio‑recorded, transcribed verbatim, and analyzed using Braun and Clarke’s six‑phase thematic analysis framework. Results Four main themes and sixteen subthemes emerged. Participants recognized substantial benefits: accelerated literature retrieval, data processing, and manuscript polishing; epistemic breakthroughs (e.g., medical image recognition, proteomic analysis, early disease prediction); and expanded research ideas. However, they expressed deep concerns, most prominently cognitive atrophy and decreased creativity (reported by 16/18 participants), factual hallucinations, patient data leakage, responsibility ambiguity, and research homogenization. Despite these risks, students did not advocate banning GenAI but instead articulated conditional acceptance with clear boundaries: acceptable uses included English polishing and code checking; unacceptable uses included generating patient diagnoses, handling identifiable clinical data, and formulating original hypotheses. Students also specified desired features of a responsible AI assistant: evidence‑based support, privacy compliance (HIPAA/GDPR‑like), accuracy and interpretability, a clear assistive role, fairness, and continuous learning. Conclusion Medical graduate students view GenAI as a double‑edged tool that offers substantial academic efficiency and epistemic gains but also poses serious risks to cognitive autonomy, accountability, and innovation. They develop mature boundary‑setting strategies and call for responsible AI designs aligned with evidence‑based medicine and ethical standards. Educators and policymakers should integrate students’ voices into curriculum and guideline development to support safe and effective GenAI use in medical graduate education.</p>