Self-Regulated Learning (SRL) is a core construct in educational sciences, describing students’ ability to activate cognitive, metacognitive, motivational, and behavioral processes aimed at achieving learning goals. SRL is articulated in phases of forethought, performance, and self-reflection, and is tightly connected with executive functions such as planning, inhibition, and working memory. While students with Special Educational Needs often experience difficulties in developing SRL competences, recent advances in Artificial Intelligence (AI) in education, adaptive systems, automated feedback, learning analytics, and intelligent tutoring systems, offer new opportunities for personalized scaffolding. However, research still lacks systematic instruments to evaluate whether and how AI functionalities effectively foster SRL. This paper addresses this gap by proposing a conceptual evaluation framework that integrates Zimmerman’s cyclical model, Pintrich’s multidimensional approach, and the Winne-Hadwin process-oriented model of SRL with updated taxonomies of AI in education that include GenAI and conversational agents. The framework operationalizes the relationship between AI functionalities and SRL phases/areas through observable indicators, concrete examples (e.g., ALEKS, ASSISTments, AutoTutor), and measurement tools that combine self-report instruments (MSLQ, MAI, LASSI) with trace-based, microanalytic protocols recommended by contemporary SRL analytics research. As an initial conceptual proposal, the framework requires empirical validation through controlled studies that assess construct validity, temporal stability, and cross-context generalizability. Despite these limitations, it contributes to theory and practice by systematizing analysis of AI’s impact on SRL, guiding researchers in designing validation studies, and offering educators and developers a structured guide for responsible AI adoption in inclusive educational contexts.