The accelerating integration of artificial intelligence into religious inquiry generates two distinct but related crises: an epistemological crisis, as algorithmic systems function as unauthorised exegetes operating beyond scholarly methodological constraints; and a structural crisis, as these systems redistribute interpretive authority in ways that systematically marginalise non-Western, non-secular epistemologies. This paper asks whether AI models replicate patriarchal weaponisation of sacred narrative or impose Western feminist frameworks as surrogate orthodoxies. Using a comparative theological framework, it examines two divergent systems Claude (Anthropic) and NurAI (Asia-based, trained by several key religious bodies) navigating female subjectivity and divine decree in Surah Maryam (Quran 19), with particular attention to Maryam's ordained silence in Verse 26 and verses 16-21, selected for their interpretive density and scholarly contestation. Findings reveal a paradox transcending the binary of bias versus accuracy: both models either secularise sacred narrative through a Western-centric lens or flatten Islamic scholarly pluralism into what this paper terms digital orthodoxy a form of AI-generated interpretive inequality structurally undetectable by existing regulatory instruments. This undetectability is not accidental. Current AI ethics frameworks the EU AI Act, UNESCO's Recommendation, and the NIST RMF are theologically illiterate by design. Because the failure is epistemological rather than procedural, no technical amendment can resolve it; the foundation itself must be reconceived. This paper proposes Digital Adab (Propriety in Code), grounded in Karamah (inherent human dignity) and operating through comparative virtue ethics, not as a corrective supplement but as a Sacred-Centred Justice model, an alternative epistemological foundation for religiously-literate AI governance.