The integration of artificial intelligence (AI) into educational systems provides numerous opportunities for personalized learning while also creating concerns about ethical data usage. This study investigates students’ perceptions of how educational AI systems should handle academic, personal, and demographic information. We survey college students from the United States (N = 128) to explore their views on permissible data types, performance thresholds for using demographics, and shifts in opinion when personally impacted. We found that most students reject the usage of political affiliation, yet many still accept using some sensitive information in models such as demographics (p < .005, 95% CI [0.763, 0.893]). Personal stakes in the system result in shifts in data-sharing preferences. Additionally, some individual differences, such as gender, correlate with willingness to share sensitive information. These findings emphasize the importance of involving students as stakeholders to design ethical and equitable educational AI systems.