<![CDATA[Personalization without autonomy is not necessarily education. Artificial intelligence (AI) can adapt support to individual learners, yet the same system may prompt planning and reflection in one classroom while supplying answers in another. This conceptual paper addresses that divergence through the Teacher-Mediated AI Regulation (T-MAIR) framework for K–12 education. T-MAIR integrates cyclical self-regulated learning, metacognitive monitoring and control, scaffolding and co-regulation, and digital-equity research with recent school-based evidence on adaptive and generative AI. A structured conceptual synthesis, updated through August 2026, treats empirical studies as mechanism-revealing anchor cases rather than as a pooled effectiveness estimate. Recent meta-analytic evidence reports a moderate overall association between AI-supported self-regulated learning (g = 0.507), with stronger effects during performance than during forethought, highlighting a performance–competence divide. T-MAIR proposes that durable metacognitive growth depends on four linked conditions: preservation of learner regulatory opportunities, purposeful teacher mediation, readiness-contingent fading of support, and sufficient school capacity to sustain those practices. Five testable propositions follow: P1, higher Learner Regulatory Opportunity (LRO) predicts stronger transfer; P2, teacher mediation changes the function of identical AI support; P3, optimal scaffolding intensity depends on regulatory readiness and task difficulty; P4, school resources affect outcomes partly through mediation capacity; and P5, assisted performance can mask dependence when independent transfer is not assessed. The paper proposes the LRO index, the Frame-Constrain-Interrogate-Fade (FCIF) protocol, and a multilevel mixed-methods design for empirical testing. The defining criterion for successful AI integration is therefore not assisted task completion, but increasingly independent regulation when AI support is reduced or removed.]]>
Copyrights © 2026
