<![CDATA[This study examines the integration of artificial intelligence (AI) into Arabic language education in Islamic institutions through a multi-site mixed-methods design across Indonesia and Malaysia. Addressing the limited empirical attention to socio-technical, cultural, and religious dimensions of AI-mediated Arabic learning, the study investigates how AI shapes linguistic development, pedagogical practice, learner autonomy, and institutional decision-making. Data were collected from five educational settings: pesantren in Wonosobo and Bangil, UIN Antasari Banjarmasin, UII Darullughah Wadda’wah Pasuruan, and an Islamic institution in Kedah, Malaysia. Classroom observations, interviews, focus group discussions, surveys, digital artifacts, and learning documents were analyzed through triangulation. The findings show that AI supports Arabic learning most effectively as a micro-scaffolding tool for grammar feedback, vocabulary practice, pronunciation support, error detection, and guided interaction. However, its effectiveness depends on teacher mediation, clear learning objectives, and ethical-pedagogical boundaries. Three main constraints emerged: goal diffusion and over-reliance when AI tasks lack guardrails; uneven digital readiness and infrastructure; and persistent inaccuracies in literary, poetic, and classical Arabic texts requiring dzawq lughawī, rhetorical nuance, and religious interpretation. Comparative findings indicate that Indonesian sites adopted AI more progressively through pesantren-based filters, whereas Kedah applied a selective, needs-based approach grounded in traditional learning preferences. Theoretically, this study proposes a socio-technical model that positions AI as a cognitive accelerator within a teacher-led, value-anchored Arabic learning ecosystem. Practically, it offers principles for guardrailed prompting, ethical AI use, teacher capacity-building, and institutionally grounded AI integration.]]>
