The increasing prevalence of artificial intelligence (AI)-based learning systems unleashes new potentials in designing personalized learning experiences that enhance learning outcomes. However, prior research indicates that such systems can negatively impact engagement due to issues in human information processing. This study examines whether giving learners control over task difficulty selection in personalized learning systems can mitigate these effects. A laboratory within-subjects experiment involving 80 participants explored how control over personalized vocabulary learning affects learning performance and autonomy satisfaction. As such a control feature may lead to deeper information processing, I investigate the mediating effect of cognitive workload on the main effect. The study aims to contribute to human-AI interaction literature by shedding light on the importance of control in personalized system design and investigating its effects on cognitive workload, overall task performance, and autonomy satisfaction.