Artificial Co-Regulation: AI Tutors That Never Fade Capture Learners’ Regulation Instead of Transferring It

Conversational AI tutors operate at national scale in some school systems and are used by most university students, while field experiments show assisted performance rising as unassisted performance falls. This paper argues that such agents produce artificial co-regulation: they recruit the learner’s attachment system while lacking the feature that made human scaffolding developmental, contingent fading. The argument draws on scaffolding theory, attachment theory, self-regulated learning research, and recent evidence on attachment and dependence in chatbot use. Because a scaffold that never fades does not transfer regulation, performance improves while the agent is present but unassisted competence, calibration, and epistemic independence lag, most consequentially at reading transitions, with multilingual and first-generation university students as the central case. So “permanent partner” models of human–AI scaffolding describe a developmental end state, not a starting point: shared epistemic agency requires a learner capable of partnership, and five design and measurement standards would protect that precondition.Preprint of a manuscript submitted to Educational Psychology (Taylor & Francis), special issue “AI in Educational Psychology,” 31 October 2026; version 1, under review.