A Scoping Review of Generative AI's Dual Role as Scaffolding or Substitution in Self-Regulated Learning and Learner Autonomy in EFL Contexts

Generative artificial intelligence (GenAI) is increasingly embedded in English as a Foreign Language (EFL) instruction, with self-regulated learning (SRL) and learner autonomy positioned as key outcomes of interest. A growing evidence base reports positive effects of GenAI on self-regulatory strategies, alongside a parallel and largely separate literature warning that GenAI use may foster dependency, cognitive offloading and the externalization of self-monitoring. This scoping review maps this apparent paradox, in which GenAI appears simultaneously as scaffolding and as substitution, across the SRL and learner-autonomy literature, with particular attention to its implications for EFL learners. Following a PRISMA-ScR-informed approach, 12 empirical and review studies published between 2025 and 2026 were thematically synthesized alongside four foundational theoretical works. As a result, four thematic clusters emerged: (1) evidence of GenAI functioning as scaffolding for SRL; (2) evidence of GenAI functioning as substitution, fostering dependency and cognitive offloading; (3) design and timing features that moderate which role GenAI plays; and (4) existing reviews and meta-analyses, together with their methodological limitations. Across clusters, a consistent pattern emerges, in that nearly all studies rely on short-term or cross-sectional designs, and several recent reviews explicitly call for research measuring learners’ delayed, unsupported performance so as to distinguish durable internalization from temporary dependency. No identified study has applied a longitudinal design incorporating a structured AI-withdrawal phase to directly test this distinction among EFL learners. This review argues for such a design and outlines its theoretical and methodological implications for Vietnamese EFL contexts.