To explore the role of parents in secondary school students' persistence in human-AI hybrid learning, this mixed-method study proposes a model that integrates active and restrictive parental mediation into an established baseline model from prior research. Using structural equation modelling to analyse data from 302 students, the proposed model accounted for 66.8% of the variance in continuance intention and 22.9% in continuous learning behaviour. Interestingly, while restrictive mediation was found to be positively correlated with students' continuous learning behaviour, active mediation did not show a significant association with the behaviour. This unexpected finding prompted further investigation through in-depth interviews with 10 students. Thematic analysis revealed that many parents focused on managing online entertainment behaviour. Restrictive mediation, usually implemented through mandatory measures, was found to be more effective in shifting students' attention from entertainment to academic activities. However, many parents lacked sufficient AI literacy and did not receive adequate support from schools and teachers. This deficiency hampered their ability to proactively help their children use AI technologies for learning, thereby limiting the effectiveness of active mediation in promoting sustained educational engagement. The results provide insights for parents, technology companies and educators to optimise human-AI hybrid learning for secondary school students.