CAPE TOWN — In an era where artificial intelligence and machine learning intersect increasingly with traditional ethics, technology developers frequently navigate uncharted waters regarding religious compliance. Recently, a software engineer building video-detection technology sought formal guidance on whether the collection and processing of openly licensed photographs for machine learning models comply with Islamic law, particularly when images feature individuals without traditional Islamic attire.
Answering the query, prominent South African Islamic scholar Shaykh Irshaad Sedick issued a detailed legal clarification (fatwa) affirming that developing, utilizing, and commercializing software that detects people and objects in video feeds is permissible under Islamic law, provided that human oversight remains strictly controlled and automated filters effectively screen out indecent material.
The ruling addresses a common technical reality in modern software engineering: training computer vision models typically requires vast datasets of real-world images. The developer explained that their pipeline downloads openly licensed photographs, processes them through an automated filter to remove indecent material, and relies entirely on machine processing rather than human viewing. The resulting trained model contains only numerical weights—pure mathematical data—with no underlying image files embedded within the commercialized software.
The Principle of Human Vision and Intent
At the core of the ruling is a foundational legal distinction within Islamic jurisprudence concerning the act of looking. Citing the classical Shafi’i jurist Imam an-Nawawi in Minhaj al-Talibin, Shaykh Sedick emphasized that the traditional prohibition against viewing the nakedness or uncovered parts of unrelated individuals is inherently defined as a human act.
"Every word of that describes a human act," the scholar explained, noting that a computer program possesses neither eyes nor religious accountability. Consequently, a folder of digital files residing on a server that no human opens or reviews produces no act of looking under Islamic law.
The fatwa further draws upon classical precedents regarding reflections and likenesses. Citing rulings from renowned jurists such as Imam Ibn Hajar al-Haytami and Imam ar-Ramli in texts like Tuhfat al-Muhtaj and Nihayat al-Muhtaj, the scholar noted that viewing a likeness or a reflection—such as in a mirror—is traditionally viewed distinctly from direct human sight, provided there is no fear of temptation or desire.
While acknowledging that contemporary scholars and guidance platforms maintain caution regarding digital screens and pictures, Shaykh Sedick emphasized that the complete absence of a human viewer in the developer’s automated pipeline renders these concerns inapplicable. He advised, however, that any necessary quality-assurance checks involving human reviewers must be carefully engineered using blurred, cropped, or appropriately covered subjects to protect team members from viewing content inconsistent with Islamic modesty standards.
Digital Duplication and the Status of Image Files
Addressing the technical aspects of downloading and storing image files, the fatwa examined whether dataset collection constitutes the prohibited creation or display of images. Islamic legal texts traditionally differentiate between making animate forms and the secondary handling of such images.
Referring to commentary by Imam an-Nawawi in Sharh Sahih Muslim, Shaykh Sedick noted that the permissibility of keeping an image depends heavily on its context, distinguishing between images displayed prominently with honor, such as those hung on walls, and those treated without distinction.
Furthermore, the ruling cited a recognized position among contemporary Shafi’i scholars of Tarim—including the late Mufti Salim Baghithan—which holds that modern photography and digital duplication do not constitute the prohibited act of image-making (taswir), as the underlying technology does not bring a physical form into independent existence in the traditional sense. Duplicating digital files for machine learning datasets, stored securely on a drive without public display or human viewing, stands far removed from any prohibited veneration or display of images.
Scope of Existing Regional Fatwas
The inquiry also addressed potential conflicts with broader regulatory rulings, specifically referencing Fatwa MUI 287 of 2001 issued by the Indonesian Council of Ulema regarding pornography.
Shaykh Sedick clarified that the scope of such rulings is explicitly defined as addressing pornography and material designed to arouse desire, prohibiting the reproduction, distribution, and viewing of uncovered nakedness outside of legally justified contexts. Because the developer’s automated pipeline actively excludes swimwear, underwear, and nudity, and because no human views the residual dataset, the activity falls outside the restrictive scope of the fatwa. He advised maintaining strict automated filters and keeping private datasets confidential to respect the spirit of such regulatory guidelines.
Commercialization and the Default of Permissibility
In accordance with established legal maxims—such as the principle cited by Imam as-Suyuti in al-Ashbah wa al-Naza’ir that "the default in things is permissibility until evidence establishes prohibition"—the creation and sale of lawful software products remain fully permissible.
The resulting software model delivered to buyers contains exclusively numerical parameters rather than photographs, making the object of sale entirely lawful and clean of prohibited elements. The scholar noted that even in complex commercial scenarios where external or secondary elements might be viewed with caution, established jurisprudence often upholds the validity of the underlying contract itself, drawing parallels to classical discussions on permissible trade transactions.
Concluding the guidance, Shaykh Sedick commended the developer’s diligence in seeking lawful provisions for modern technology, affirming that existing models do not require discarding or retraining, and that developers may proceed with confidence in building and distributing their computer vision tools.