Generative artificial intelligence (AI) has become an indispensable resource in contemporary higher education, providing substantial benefits to both students and institutions. As its adoption accelerates, it is important to balance these advantages against potential risks that may arise from students’ varying levels of interaction with AI. Whereas most prior studies have focused on the favorable outcomes of AI for college students, the present research investigates its “double-edged sword” effects. Guided by social affiliation theory, a new model has been developed and empirically tested to clarify how and under what conditions the frequency of student–AI interaction influences social needs and behaviors. Longitudinal data obtained from 388 undergraduates showed that AI interaction frequency can shape prosocial behavior and problematic mobile phone use through a dual pathway involving the need for affiliation and feelings of loneliness. Furthermore, peer support moderates the indirect effect of AI interaction frequency on college students’ prosocial behavior via their need for affiliation. The results showed that peer support did not significantly moderate the indirect relationship between AI interaction frequency and problematic mobile phone use through loneliness. Overall, our study extends the framework of the social affiliation theory and provides practical insights that guide the appropriate use of AI by college students, thereby supporting the development of healthy social skills and technology engagement in the age of AI.