As AI technologies like GPT models continue to reshape various aspects of society, it is imperative to investigate the perceptions and ethical considerations of graduate students regarding GPT's use in academic settings. This mixed-method exploratory study engaged 21 graduate students through surveys and focus group interviews. The findings revealed: (1) while engagement with GPT was limited and focused, incidents of unethical use were already present; (2) after experimenting with a designated GPT website, participants showed a statistically significant improvement in their Intention to use AI in research, whereas their research self-efficacy, course satisfaction, perception of AI, necessity of AI for career development, and perceived AI future remained unchanged; (3) graduate students are more likely to use GPT for academic misconduct under time pressure and challenging coursework, exacerbated by instructors' limited AI awareness. The study advocates for clearer definitions of academic misconduct and a concerted effort toward responsible AI use.