Artificial Intelligence Colonialism: Environmental Damage, Labor Exploitation, and Human Rights Crises in the Global South

Salvador Santino F. Regilme analyzes AI’s supply chains through colonialism, political economy and human-rights scholarship. His question is who benefits from AI development and who supplies its labor, materials and ecological capacity. The article argues that concentrating benefits while exporting burdens can reproduce colonial relations even without direct territorial rule.^[1]

Two forms of unequal burden

The analysis distinguishes necroexportation, the externalization of harmful consequences onto marginalized populations and ecosystems, from necrostratification, the hierarchy of whose lives and futures receive protection. In Regilme’s account, the first describes movement of costs across unequal relationships; the second describes how benefits and vulnerability are distributed within and across societies. These are interpretive concepts, not two independently measured indices (pp. 77–79).^[1]

The Global North–South framing therefore does more than compare national access to software. It connects ownership, subcontracting and resource dependence to the power to choose whose costs count. The article relates these ideas to biopolitics, necropolitics and surveillance capitalism, synthesizing scholarship and reported cases rather than conducting a new worker survey or supply-chain experiment.^[1]

The labor behind apparent autonomy

Regilme emphasizes data labeling and content moderation: systems presented as autonomous depend on people sorting examples, judging material and preparing data. His account connects this work to subcontracting, short-term contracts, platform monitoring and limited worker protection. Ghost Work captures a related problem of essential labor disappearing from the product’s public story (pp. 79–83).^[1]

The argument is about employment relations as well as compensation. A platform can distribute small tasks globally while leaving workers with unpredictable demand and little influence over working conditions. Regilme interprets the contrast between that vulnerability and the rewards of owners and senior technical staff as a hierarchy of recognition and security. His particular wage and prevalence figures come from other reports; this article does not establish a representative estimate for all AI workers.^[1]

Environmental effects across a lifecycle

The paper separates direct impacts of producing, transporting, operating and retiring computing resources from indirect effects of AI applications in sectors such as mining and manufacturing. An efficiency improvement can also increase total activity, so reducing the resources needed for one task need not reduce the aggregate burden. The article acknowledges the difficulty of separating AI-specific impacts from broader information-and-communication technology activity (pp. 83–86).^[1]

Regilme links energy and cooling demand, material extraction and climate vulnerability to his account of externalized costs. That is a political-economy interpretation of varied evidence, rather than a lifecycle calculation tracing a particular model to a particular community’s harm. Mining, data-center and emissions figures therefore require checking their underlying scope before attributing them specifically to AI.^[1]

What a rights-based response would cover

The conclusion connects present workers’ dignity and privacy with environmental protection and future generations’ interests. It also acknowledges beneficial uses of AI, including health and disaster-response applications. The proposed response is governance that accounts for the people and ecosystems supporting a system, rather than assuming that technical innovation alone distributes its gains fairly (pp. 86–87). These are normative demands, not interventions tested for effectiveness.^[1]

For Social Alignment, the contribution is to widen the affected group beyond a model’s user or owner. A system can satisfy a client’s instructions while its production conditions remain objectionable. Regilme’s analysis asks who can contest those conditions and whether the costs of providing useful AI are visible in the first place.^[1]

The article appears in SAIS Review of International Affairs, volume 44, issue 2, pp. 75–92, labeled Fall–Winter 2024 with publisher copyright 2025. Leiden’s catalog gives February 9, 2025; that record does not independently establish the first online-release day. The Leiden and ResearchGate copies represent the same DOI-bearing article, not separate studies.^[1] See Salvador Santino F. Regilme.