From Learner Modeling to Human–AI Co-Regulation: A Systematic Scoping Review of Adaptive Generative AI in Personalized Learning

Abstract* Background Adaptive generative artificial intelligence (GenAI) is increasingly used to personalize learning, yet evidence remains fragmented across system design, human–AI co-regulation, self-regulated learning (SRL), sustained behavioral outcomes, and ethics. This systematic scoping review mapped these dimensions in a single synthesis. Methods We searched Dimensions, ERIC, IEEE Xplore, Lens, PubMed, Scopus, and Web of Science on 5 May 2026. Of 3,084 records identified, 95 journal articles published between 2023 and 2026 met the inclusion criteria. Evidence was synthesized using descriptive mapping, form–function co-occurrence analysis, and qualitative comparative synthesis. Results Adaptive GenAI was most commonly instantiated as conversational tutors, generative content personalizers, and hybrid feedback systems, with adaptive feedback, content adaptation, diagnostic analysis, scaffolding, and progress tracking emerging as the dominant functions. Conceptualizations of human–AI co-adaptation were less mature: most studies framed AI primarily as a responsive regulator, with feedback-loop acceleration and metacognitive prompting more frequently articulated than internalization, behavioral reinforcement, or social modeling. Empirical effects were concentrated in proximal SRL outcomes, particularly forethought and performance, while evidence for self-reflection, affective regulation, transfer, stable self-efficacy, persistence, and learning habits was much thinner. Ethical evaluation likewise remained uneven, emphasizing feedback quality, pedagogical validity, well-being, autonomy, motivation, inclusivity, and agency, while transparency, data minimization, scaffold fading, developmental suitability, manipulation risk, and ethical nudging were addressed far less often. Conclusions Adaptive GenAI is more mature in technical personalization and proximal SRL support than in reciprocal co-regulation, durable behavior change, and integrated ethical governance. Future research should test whether regulatory gains persist beyond immediate AI support and develop stronger frameworks for governing learner influence and dependency.