<![CDATA[This study examines authority and attribution in Qur’anic interpretation mediated by artificial intelligence. The phrase fī sabīlillāh in Q 9:60 serves as a test case because its meaning has been disputed since the earliest period, its transmission is layered, and its implications extend to zakat distribution. The same interpretive question was submitted to seven mainstream language models and assessed through tafsīr bi al-ma’ṡūr, understood as an evaluative hermeneutic centered on traceable attribution, isnād, tarjīḥ, and fidelity to sources. Interpretations by al-Ṭabarī, al-Qurṭubī, and Ibn Kaṡīr provided the basis for an analytic rubric. Two raters independently evaluated 28 model responses and reconciled their judgments through consensus, producing 248 audited error findings. Although the models generally identified the three exegetes and their works, they performed poorly in preserving isnād, narration structure, citation accuracy, and attribution. Unsupported additions and shifts of meaning dominated the errors. Higher-performing models produced more polished answers but did not eliminate source-level failures. The findings indicate that greater model capacity alone cannot secure credible machine-mediated tafsir; verification must restore the attributional discipline of tafsīr bi al-ma’ṡūr.]]>
Copyrights © 2026
