Teacher Mediation and the Contingent Promise of Generative <scp>AI</scp> in Mathematics Education

ABSTRACT Background The integration of generative AI (GenAI) into education, specifically into teaching and learning, offers potential for personalised learning, but at the same time, it poses a challenge to whether it fosters student agency. Moreover, it also threatens to exacerbate entrenched inequalities, which have the potential to create new barriers for most of the learners. A gap exists in understanding how GenAI can be rethought as a dynamic scaffold that promotes cognitive struggle for conceptual understanding while actively advancing equitable participation. Objective The study aims to critically analyse how secondary school mathematics teachers implement GenAI for adaptive scaffolding. Moreover, this study aims to investigate the pedagogical and ethical tensions, particularly concerning equity and power dynamics, that emerge from using GenAI during classroom interactions. Methods This study employed a qualitative research design. Data were collected over an eight‐week intervention, which included in‐depth teacher interviews ( n = 10), student focus groups ( n = 24), and classroom observations in two public secondary schools purposively selected for their contrasting socioeconomic and technology integration profiles. Results Based on the results and findings, five distinct teacher‐mediated GenAI scaffolding practices that support equitable knowledge construction were identified. Findings indicate that strategies such as prompting for problem formulation through AI‐generated inquiry and facilitating critical comparison of AI‐generated outputs were associated with deeper student engagement and reasoning within the conditions of this study, particularly when teachers exercised intentional critical mediation. Conversely, unmediated use of GenAI amplified significant challenges, such as a redistribution of epistemic authority, a tangible “algorithmic divide” in AI performance between well‐resourced and under‐resourced schools, and risks due to the absence of how data should be handled. The technology consistently amplified existing pedagogical conditions, which are both effective and inequitable. Conclusion The promise of GenAI is strictly contingent on a commitment to equity. This means that its outcomes are determined not by the GenAI itself but by pedagogical intentionality and systemic support. Unexamined adoption poses risks that worsen disparities for marginalised students and reinforces passive learning. Thus, by proposing a framework for equitable integration centred on the teacher's role as a critical mediator, this expertise serves as a model for sustainable pedagogical practice in the digital age.