Mohammad S Razai is an NIHR Clinical Lecturer in Primary Care at the University of Cambridge. He is on Bluesky and LinkedIn.
Alisina Mahmood is a Research Fellow at the University of Tokyo.
Seyed Aria Nejadghaderi is a Research Fellow at the Kerman University of Medical Sciences.
Hossein Dabbagh is an Associate Professor of Philosophy at Northeastern University London and Philosophy Tutor at the University of Oxford.
Artificial intelligence (AI) is moving into population-level decisions about health entitlements, access, and payment. In some low- and middle-income countries, these systems are being adopted amid workforce shortages, weak infrastructure, donor dependency, limited regulation, and socioeconomic inequalities. For people with limited control over the conditions that affect their health, AI adds another layer of decision making that is difficult to understand or challenge. The central moral concern is not whether these systems work, but whether they expand or restrict people’s ability to access care, contest decisions, and shape the conditions under which health decisions are made.
Automated decision making offers a warning for health systems. Poverty and social need can be turned into risk scores, surveillance triggers, and eligibility decisions.1 Search algorithms can reproduce racial and gendered hierarchies even when they appear neutral.2 Large-scale predictive models could distribute harm through opaque, byzantine rules that affected people cannot easily understand or challenge.3
The right to health provides AI governance with a clear legal and ethical test by shifting attention from innovation in the abstract to the concrete conditions under which people can access care. The United Nations treaty on economic, social, and cultural rights recognises everyone’s rights to the highest attainable standard of physical and mental health.4 It sets out four linked requirements: health facilities, goods, and services must be available, accessible, acceptable, and of good quality.5 A tool that improves diagnostic speed but depends on smartphones, a stable internet connection, or paid data may widen access for some while excluding others. A risk score that guides the allocation of scarce resources may also worsen inequity if trained on records that undercount those least able to access care.
For people with limited control over the conditions that affect their health, AI adds another layer of decision making that is difficult to understand or challenge.
The World Health Organization guidance on large multi-modal models warns against ‘technological solutionism’, unequal access, system-wide bias, and dependence on models maintained for high-income settings. Safeguards include transparency about training data, evidence of safety and performance, human rights impact assessments, public participation, inclusive pricing and language access, post-release audit, and routes to redress.6,7
These safeguards are insufficient unless governance returns to the question of power. Who defines the problem? Whose data are used? Who owns the tool? Who can challenge decisions? Who benefits financially? These are ethical, not merely implementation, questions because such systems can decide whose needs become visible, whose suffering is translated into data, and whose claims are treated as credible.
Kenya shows the risks of AI health-financing decisions. A 2026 investigation found that Kenya’s Social Health Authority used a predictive means-testing algorithm to estimate household income and set insurance contributions. Poorer households were reportedly overcharged, while wealthier households were undercharged.8 Whatever the technical rationale, an opaque algorithm that affects whether people can afford care raises clear right-to-health concerns.
… safeguards are insufficient unless governance returns to the question of power. Who defines the problem? Whose data are used? Who owns the tool? Who can challenge decisions? Who benefits financially?
Data extraction is another risk. Populations in the Global South — a diverse grouping with varied health systems and regulatory capacity — may supply clinical data for systems designed and monetised elsewhere. This fuels concerns about ‘data colonialism’: the extraction and control of human data by big tech companies and governments for profit.9 Unequal control over data, infrastructure, and analytical capacity can reproduce historical patterns of dependency. Critics of ‘algorithmic colonisation’ in Africa similarly warn that externally imposed technologies, corporate agendas, and data practices can perpetuate colonial exploitation.10
Before deployment, governments and tech companies need to demonstrate local need and validity, and publish evidence on safety, bias, language performance, failure modes, and real-world use. Procurement should require a human rights impact assessment, covering discrimination, privacy, participation, clinical safety, and affordability. Affected patients and communities must have clear routes to explanation, appeal, and remedy.6,7 Contracts should include suspension and withdrawal clauses, so unsafe or inequitable systems can be stopped without locking health systems into continued use.
Participatory practices, including co-production and co-creation, should be applied from the bottom up. Patients, practitioners, disability groups, women’s organisations, and rural and minoritised communities should be involved in the design brief, data rules, evaluation metrics, and scale-up decisions. Local researchers and public institutions should retain genuine control over governance and evaluation. Responsible AI requires Global South leadership, with tools judged by health needs and public value rather than market scale.
Data extraction … fuels concerns about ‘data colonialism’: the extraction and control of human data by big tech companies and governments for profit.
AI will change access to care, the distribution of resources, and the visibility of communities within health systems. Its place in healthcare in the Global South should therefore be determined through human rights standards, public deliberation, and accountable governance.
References
1. Eubanks V. Automating inequality: how high-tech tools profile, police, and punish the poor. New York: St Martin’s Press, 2018.
2. Noble SU. Algorithms of oppression: how search engines reinforce racism. New York: New York University Press, 2018.
3. O’Neil C. Weapons of math destruction: how big data increases inequality and threatens democracy. New York: Crown, 2017.
4. United Nations General Assembly. International covenant on economic, social and cultural rights. Geneva: United Nations, 1966. https://www.ohchr.org/en/instruments-mechanisms/instruments/international-covenant-economic-social-and-cultural-rights (accessed 21 Sep 2026).
5. Office of the High Commissioner for Human Rights. General Comment No. 14: The right to the highest attainable standard of health. New York: United Nations, 2000. https://www.ohchr.org/sites/default/files/Documents/Issues/Women/WRGS/Health/GC14.pdf (accessed 22 Sep 2026).
6. World Health Organization. Ethics and governance of artificial intelligence for health: guidance on large multi-modal models. Geneva: WHO, 2024. https://www.who.int/publications/i/item/9789240084759 (accessed 22 Sep 2026).
7. World Health Organization. Ethics and governance of artificial intelligence for health. Geneva: WHO, 2021. https://www.who.int/publications/i/item/9789240029200 (accessed 22 Sep 2026).
8. Mukami P, Kirigia J, Geiger G, et al. Hiding behind AI: how Kenya used a predictive algorithm to transfer health costs onto the poorest. Lighthouse Reports, 2026. https://www.lighthousereports.com/investigation/hiding-behind-ai/ (accessed 22 Sep 2026).
9. Couldry N, Mejias UA. Data colonialism: rethinking big data’s relation to the contemporary subject. Telev New Media 2019; 20(4): 336–349.
10. Birhane A. Algorithmic colonization of Africa. SCRIPTed 2020; 17(2): 389.
Featured photo by Steve A Johnson on Unsplash.
