This systematic literature review analyzes 17 peer-reviewed studies published between 2023 and 2025 to examine how artificial intelligence tools function as pedagogical instruments for developing students' creativity. The review draws on sources indexed in Scopus, Web of Science, ERIC, and CCES databases. The analysis reveals that AI-assisted instruction produces measurable gains in creative performance (g = 0.867 for learning outcomes; g = 0.457 for higher-order thinking), yet simultaneously creates what we term the "creativity paradox": learners produce higher-quality work but report reduced personal engagement, cognitive effort, and intrinsic motivation. Interpreted through Self-Determination Theory and constructivist pedagogy, this paradox reflects a tension between competence support and autonomy undermining. The review identifies four pedagogical conditions that mediate this effect: scaffolded AI interaction grounded in the zone of proximal development, explicit prompt engineering instruction as a metacognitive skill, blended AI-human feedback within a formative assessment framework, and task design requiring personal investment and critical reflection. Academic integrity concerns appear in 64.7% of studies, and over-reliance on AI is noted in 52.9%. The findings demonstrate that pedagogical design - not the technology itself - determines whether AI develops or diminishes students' creativity. The review contributes to educational theory by framing AI-enhanced creativity within constructivist and motivational frameworks, and offers evidence-based recommendations for educators. Research gaps in longitudinal designs and diverse educational contexts point to directions for future investigation.