What question did this study set out to answer?

The aim is to propose a developmental framework for Generative AI integration in educational contexts across primary, secondary, and tertiary levels.

June 11, 2026Open Access

Teaching With, Not Around A Developmental Framework for Generative AI Integration in Primary, Secondary, and Tertiary Education (2023–2026)

Key Points

The aim is to propose a developmental framework for Generative AI integration in educational contexts across primary, secondary, and tertiary levels.
Synthesized empirical research from 2023 to 2026 regarding cognitive offloading and developmental neuroscience.
Evaluated the impact of generative AI on learner tiers through the 4D framework.
Analyzed existing AI literacy frameworks and their relevance to educational policies.
GenAI exposure should be adapted based on developmental tiers rather than uniformly applied.
Constitutional AI methodologies provide a stronger foundation for educational model selection than non-transparent systems.
The educator's role is changing to focus more on cognitive coaching, necessitating corresponding training adjustments.

Abstract

Generative artificial intelligence (GenAI) is the fastest-diffusing general-purpose technology in modern educational history. Between November 2022 and early 2026, frontier large language models moved from research curiosity to ubiquitous student tool in three academic cycles. Education systems did not. This paper argues that the dominant institutional postures, prohibition, unregulated adoption, and surveillance-based compliance, are all developmentally illiterate, and proposes a four-stage cognitive framework (Delegation, Description, Discernment, Diligence; hereafter 4D) calibrated against the neurodevelopmental realities of primary, secondary, and tertiary learners. Drawing on a synthesis of empirical work from 2023 to 2026 on cognitive offloading (Kosmyna et al., 2025; Gerlich, 2025; Fan et al., 2025; Gilbert, 2024), the developmental neuroscience of prefrontal maturation under digital exposure (Hutton et al., 2022; Hinze et al., 2025; Lewin et al., 2025; Gascón-Cánovas et al., 2025), the EU and OECD AI literacy frameworks (European Commission UNESCO, 2024), and the forthcoming PISA 2029 Media and AI Literacy assessment (OECD, 2026), the paper makes four claims. First, GenAI exposure should be structurally asymmetric across developmental tiers, not uniformly permitted or uniformly forbidden. Second, model selection in regulated educational environments is a non-trivial pedagogical choice; methodology-transparent systems such as those trained under Constitutional AI protocols (Bai et al., 2022) present a more defensible foundation for institutional deployment than systems where alignment methodology is undisclosed. Third, the educator's role is shifting from content-deliverer to cognitive coach, and educator training must shift correspondingly. Fourth, the international policy environment is converging rapidly around four-domain literacy models (engage, create, manage, design) which create both an obligation and an opportunity for institutional reform. The paper concludes with operational guidance for institutions across the 2026–2030 horizon, a limitations section, and an agenda for future empirical research. Keywords: generative AI · AI literacy · cognitive offloading · developmental neuroscience · educational technology · 4D framework · Constitutional AI · AI ethics in education · adolescent executive function · OECD AILit framework · PISA 2029 · protection-by-design pedagogy

Read Full Paperexternally

Mark Helpful

Bookmark

Relay

View Full Paper