Algorithmic riba and algorithmic Shariah non-compliance: risk mechanisms and a proposed governance protocol for Islamic banks

Abstract Machine learning (ML) systems are increasingly used in the credit, pricing, and collections decisions of financial institutions, including Islamic banks, yet Shariah oversight remains primarily contract-centred. This conceptual paper identifies three distinct pathways of algorithmic Shariah non-compliance. Algorithmic riba, used narrowly, concerns profit-rate models whose optimisation objective may reproduce the time-value logic prohibited by riba; conventional benchmarking of murabahah (cost-plus sale) margins is not itself impermissible. Algorithmic injustice concerns credit-scoring outcomes that may violate the ‘adl (justice) obligation, and algorithmic gharamah (penalty) concerns collections systems that may impose charges beyond ​ta’widh (compensation for documented loss). To support model-level review, we propose the Shariah Algorithmic Compliance Screen (SACS), a five-component protocol grounded in Maqasid al-Shariah and offered as a first-iteration design proposal. Model owners disclose use cases, training data, objective functions, and outputs; independent technical validators test the relevant model evidence; and Shariah Supervisory Boards (SSBs) determine compliance. Explainable artificial intelligence (XAI) methods are used descriptively, while established fairness diagnostics (subgroup calibration, error-rate comparison, and disparate-impact measurement) provide separate evidence on outcome equity. The three mechanisms are derived analytically from doctrine and the ML literature and are stated as conditional hypotheses. The study completes the construction stage of the design: SACS is conceptually grounded and operationally specified, and its components adapt existing artificial intelligence governance practice to jurisprudential review. SACS has not yet been evaluated. Practitioner evaluation, particularly in Small Island Developing States such as the Maldives, is the next stage and will assess intelligibility, implementation burden, and practical usefulness.