Artificial intelligence (AI) tutors can generate immediate and personalized feedback, yet their use in Islamic Religious Education (IRE) raises a distinctive problem: feedback on akhlak is not merely a prediction of answer correctness but a pedagogical intervention that may affect moral agency, dignity, religious understanding, and relationships. This conceptual design study develops a maqāṣid al-sharīʿah-based explainable AI framework for evaluating akhlak-related learning while strengthening students’ critical thinking. An integrative, criterion-based review was conducted on a purposive corpus of 50 sources: 40 peer-reviewed journal articles and 10 books, policy instruments, or technical standards. The literature was coded across five domains: AI in education, explainable AI, formative feedback, Islamic education and maqāṣid, and critical thinking and AI ethics. The synthesis produced the MAQ-XAI Akhlak Tutor, whose novelty lies in dual-loop explainable ethical feedback. The epistemic loop makes visible the learning objective, evidence, rubric dimension, reasoning gap, and uncertainty, while the maqāṣid-ethical loop explains the protected good, possible harm, dignity requirements, and need for human oversight. The framework also introduces a moral-agency handoff, five feedback traces, seven akhlak–critical thinking dimensions, and risk-based escalation that prohibits autonomous judgments in sensitive theological, psychosocial, identity, or safety-related cases. Results suggest that explainability in IRE should be evaluated not by transparency alone but by source fidelity, pedagogical usefulness, trust calibration, dignity preservation, and the quality of student revision. The framework offers a testable design for national Islamic education contexts and provides practical requirements for teachers, curriculum developers, policymakers, and educational technology providers. Empirical validation through expert review, classroom trials, and comparative studies is required before deployment.