Trust, mistrust, and the promise of AI in genomics for African populations Authors: Amadou Gaye, Vence L. Bonham, Tesfaye B. Mersha Published: 2026 ABSTRACT SummaryArtificial intelligence (AI) is rapidly reshaping genomic medicine, yet its benefits remain unevenly distributed due to the profound under-representation of African populations in genomic datasets and persistent legacies of mistrust. These structural gaps undermine model validity, amplify bias, and limit clinical utility across the world’s most genetically diverse populations. Building trustworthy AI for genomics in African populations requires transparent and interpretable systems, equitable data generation led by African institutions, culturally grounded governance, and rigorous population-specific validation. Advancing these principles is essential to ensure that AI reduces, rather than reinforces, global genomic inequities. Introduction Artificial intelligence (AI) has rapidly become a transformative force in genomics, promising to refine disease prediction, improve pharmacogenomics, and accelerate therapeutic discovery.1,2,3 Yet despite these advances, AI’s integration into genomics exposes a deep and often overlooked fault line: trust. Nowhere is this more evident than in the context of African populations, who harbor the greatest genomic diversity globally but remain under-represented in genomic datasets.4,5 Most genetic studies have historically focused on populations of European descent despite this group representing only ∼about 10%–16% of the global population.5,6,7,8 By contrast, only ∼1.1% of participants in genome-wide association studies are of African ancestry.9 Notably, recent policy guidance from major professional societies has begun to explicitly acknowledge that the equitable implementation of genomics, and the AI systems increasingly layered upon it, depends fundamentally on trust, data representativeness, and responsible stewardship across populations.10 Despite its rapid and widespread adoption in everyday settings and health care, AI poses risks to health equity because of its reliance on large datasets for training and model development. These datasets often reflect only information from patients who are able to access the healthcare system and can perpetuate the biases and structural inequities that pervade society. When it comes to Africa, the challenge is not merely scientific. It is cultural, ethical, historical, and increasingly digital. As AI becomes more deeply embedded in genomic science, the tension between trust and mistrust threatens to shape the future of health equity on the continent. The question now becomes whether AI amplifies or closes existing gaps. AI in African genomics is a double-edged sword. While it offers enormous opportunities in accelerating disease discovery, advancing pharmacogenetics, enhancing precision medicine, and enabling data-driven public health innovation, it also introduces significant risks. The rapid expansion of AI highlights structural vulnerabilities: the lack of African genomic data in AI reference sets, limited local computational capacity, and the possibility that foreign-developed technologies may deepen inequality or contribute to digital colonization.11,12 The roots of mistrust: Representation, exploitation, and historical memory Despite representing the oldest and most diverse human lineages, African populations contribute less than 2% of the world’s genomic reference data. Models trained primarily on European ancestry datasets reflect a structural imbalance in global genomics rather than a marginal limitation.13 This imbalance results in incomplete representation of genetic variation, weaker imputation reference panels, reduced transferability of polygenic risk scores, and increased uncertainty in variant interpretation. These limitations are compounded by differences in environmental exposures, health systems, and disease prevalence, leading to model miscalibration when applied outside their development populations.14,15 Because these models are typically trained on data from over-represented groups, they may overfit dominant patterns, which in turn can reinforce existing medical biases and fail to recognize disease or disease-specific outcomes in people from under-represented groups, yielding less accuracy or potential harm due to wrong recommendations for these populations. Such failures can be obscured by the metrics used to claim model validity. Methodological guidance has shown that widely reported performance measures, such as accuracy, F1 score, and related classification metrics, can remain high even when models are clinically misleading, particularly when applied outside their development populations.16 In settings characterized by population shift, including many African contexts, these metrics may mask miscalibration and decision-relevant error rather than reveal it. The question of how much African genomic data are needed cannot be reduced to a single threshold. Requirements vary depending on the intended application. For common variant discovery and polygenic risk prediction, large ancestrally diverse cohorts are necessary to capture allele frequency variation and linkage disequilibrium patterns. For rare variant interpretation, deep sequencing and phenotype-rich datasets across multiple African populations are essential. For clinical AI systems, representativeness must extend beyond ancestry to include geography, care settings, environmental exposures, and social determinants of health. This concern is not hypothetical. Large clinical and population studies summarized in recent American Heart Association (AHA) policy guidance demonstrate that lack of ancestral diversity directly contributes to variant misclassification, unreliable risk prediction, and downstream clinical error.10 When communities see tools failing to reflect their lived biology, mistrust is not only rational, it is inevitable. But the deeper roots lie in history. Extractive research practices, lack of benefit sharing, and perceived exploitation of biological samples have eroded community confidence in genomics. The absence of African data in global AI development reinforces the perception that new technologies are simply the next iteration of externally driven agendas. Without honest acknowledgment of these legacies, trust cannot be rebuilt. Black-box science and cultural disconnect Some AI approaches, particularly complex deep learning and foundation models, can function as relatively opaque systems in which the relationship between input data and predictions is not easily interpretable. However, not all AI applications in genomics fall into this category. Many widely used methods incorporate interpretable statistical frameworks, feature attribution techniques, or can be combined with domain knowledge and human curation. Nonetheless, ensuring appropriate levels of interpretability, transparency, and auditability remains essential, particularly in clinical settings.17 While interpretability challenges are increasingly addressed through explainable AI methods and regulatory oversight in high-resource settings, even limited opacity may represent a greater barrier in contexts where historical mistrust and data under-representation are more pronounced. In Africa, where mistrust is already high, algorithmic opacity becomes an obstacle to adoption. Trust requires interpretability. Many AI tools are developed primarily in English and are culturally aligned with Western contexts, which can limit their accuracy, interpretation, and accessibility for users in non-English-speaking regions.18 In addition, most AI models exclude environment-specific social determinants of health in the training models, which could affect their outputs especially in minoritized populations who are affected by these risk factors, and the results of algorithms must be interpreted with caution. Recent policy statements emphasize that explainability is essential in clinical genomics, and AI-driven models must be interpretable, population validated, and clinically auditable to be ethically deployable, particularly in contexts where trust is fragile.10 Clinicians and researchers must be able to trace how an AI model reaches its conclusions. Equally important, AI systems must incorporate cultural, environmental, and lifestyle factors that meaningfully shape health in African contexts. Models built exclusively on Western data and assumptions fail to capture African realities, widening both scientific and social distance. The digital information landscape: Fuel for trust or misinformation Internet connectivity has expanded rapidly across the African continent, driven in part by the widespread availability of low-cost smartphones that have increased access to digital infrastructure. This connectivity brings unprecedented access to information but also exposes communities to misinformation about genetics, ancestry, AI surveillance, or fears of biological exploitation. This situation creates a paradox in which AI systems risk failing African ancestral populations due to insufficient data, while mistrust simultaneously hinders the collection of the data required to address that very gap. Policy frameworks increasingly recognize that transparency around data use, AI decision-making, and downstream applications is a core mechanism for maintaining public trust because miscommunication can amplify fear and resistance.10 Transparent, culturally aware communication through trusted channels is now as essential as scientific rigor. At a time when viral misinformation and scientific racism continue to erode public trust in genetics, the unchecked rise of opaque AI systems risks amplifying skepticism, particularly among historically marginalized populations. Prioritizing AI frameworks engineered within Africa can help counter this distrust by anchoring technological innovation within local accountability structures and community engagement. By investing in African-built solutions and global partnerships, advances in AI can be connected meaningfully to local context, laying the foundation for a self-sustaining innovation ecosystem across the continent. Ethics, sovereignty, and community expectations African societies often conceive identity and heritage in communal terms. Genomic data are therefore not simply personal; they carry symbolism tied to ancestry, lineage, and spiritual values. Western approaches to consent and governance frequently fail to account for these cultural considerations. Using AI in genomics medicine poses ethical and legal concerns, such as patient privacy, data protection, informed consent, and liability concerns.19 Beyond these established concerns, AI systems introduce additional risks of algorithmic bias, large-scale data aggregation vulnerabilities, and uneven regulatory oversight across low-resource settings. Without adequate safeguards, AI tools trained on non-representative datasets may encode structural inequities directly into clinical decision systems, further widening the digital divide and reinforcing disparities rather than reducing them.11,12 Recent policy frameworks, including the AHA guidance on equitable genomic research,10 emphasize that ethical practice requires meaningful community engagement, transparent data governance, and respect for collective interests.16 These principles resonate strongly with African models of communal consent and shared data stewardship. For AI-enabled genomics to be ethically and socially acceptable, it must be embedded within frameworks that uphold community ownership of data, ensure culturally appropriate consent processes, guarantee equitable benefit sharing, and preserve local decision-making authority. In the absence of these safeguards, even the most technically robust innovations risk limited adoption or outright rejection by the communities they aim to serve. Infrastructure and autonomy as foundations of trust Gaps in computational infrastructure, cloud access, and AI expertise can create dependencies on external institutions and tools. Without efforts to improve infrastructure and affordability, AI will widen existing health inequalities rather than narrow them. Underserved communities will find themselves further marginalized where innovation bypasses those with the greatest need. This dynamic can reinforce patterns of scientific neocolonialism, whereby biological samples and data are extracted, analyzed, and monetized, while African researchers and institutions have limited control over interpretation or downstream benefits. Trust emerges when African scientists, institutions, research partnerships, and governments take the lead in model development, data analysis, interpretation, and validation efforts. African governments play a central role in ensuring the sustainability of AI-driven genomics. This includes establishing long-term funding mechanisms for genomic data generation, investing in national biobanks and secure computing infrastructure, supporting workforce development in genomics and AI, and implementing regulatory frameworks for data protection, validation, and accountability. External funding can catalyze progress but should complement, not replace, domestic investment. Sustainable impact requires co-leadership, technology transfer, and retention of analytic capacity within African institutions. These realities underscore a central truth: African genomics deserves African-built AI solutions. Developing AI tools within Africa cultivates regional talent and produces context-driven models that reflect the biological, cultural, and environmental realities of African populations. A strong local AI ecosystem reduces dependency on external technologies and ensures that algorithmic insights remain relevant, interpretable, and trustworthy for the communities they serve. Investment in local capacity is therefore not auxiliary; it is central to building a trustworthy AI ecosystem. Validation: The line between promise and harm AI tools generally perform well in the populations on which they are trained. However, their accuracy drops substantially when applied to non-reference African genomes. Without developing and rigorously validating these models in African ancestry cohorts, clinical deployment risks misclassification, missed diagnoses, and inaccurate risk stratification, ultimately undermining their utility and potentially widening existing health disparities.20,21. Population-aware validation must be a prerequisite, not an afterthought, led by African researchers and institutions. From a methodological perspective, misclassification is not merely a technical nuisance but a signal of contextual mismatch. Predictive models may retain good discrimination while systematically over- or under-estimating risk when applied in populations with different epidemiology, care pathways, or baseline risk structures. Such “miscalibration” can directly translate into overtreatment or undertreatment, reinforcing the need for validation strategies that are grounded in local populations rather than assumed portability.16 There is also the issue of contextual bias. Since AI is being developed and deployed extensively in high-income countries, and because the data utilized to create AI systems is highly tied to the context of use, the adoption of such systems in Africa may result in contextual bias.19 A path forward: Principles for trustworthy AI in African genomics A sustainable framework for trustworthy AI in African genomics must be grounded in principles that ensure scientific rigor, equity, and community partnership. First, intentional and sustained inclusion of African genomic diversity is essential for building representative datasets and mitigating algorithmic bias. Second, transparency through interpretable and explainable AI models is critical for fostering trust among clinicians, researchers, and communities. Third, governance structures must prioritize African-led ethical oversight and data sovereignty to ensure that genomic resources are stewarded in alignment with local priorities. Fourth, culturally aware, evidence-based communication strategies are needed to counter misinformation and support informed decision-making. Fifth, strengthening capacity through local infrastructure and AI expertise will enable African institutions to lead innovation rather than rely on externally developed systems. Finally, rigorous validation using African-led evaluation frameworks is required before any AI-driven genomic tool is deployed in clinical settings. Together, these pillars outline a comprehensive path toward responsible, equitable, and impactful AI in African genomic medicine. As shown in Figure 1, trust serves as the critical link between addressing historical and technical barriers and realizing the equitable benefits of AI-enabled genomics in African populations. The framework highlights current challenges, including the underrepresentation of African-led data, models, and validation, as well as historical and structural drivers of mistrust such as exploitation, poor model performance, and lack of transparency. It emphasizes guiding principles, including equity, African leadership, community governance, and transparency. Together, these elements contribute to the development of equitable and trustworthy AI for African populations.Figure 1Conceptual framework illustrating the pathway to building trust in genomics and AI for African populationsTrust is the foundation of equitable AI in African genomics. Addressing historical, technical, and governance challenges through transparency, validation, community engagement, and African leadership can enable trustworthy AI and improve genomic medicine for African populations. Conceptual framework illustrating the pathway to building trust in genomics and AI for African populations Trust is the foundation of equitable AI in African genomics. Addressing historical, technical, and governance challenges through transparency, validation, community engagement, and African leadership can enable trustworthy AI and improve genomic medicine for African populations. Ultimately, trustworthy AI should be judged not by headline performance metrics but by demonstrated clinical utility.16 Decision-analytic evaluations that explicitly account for the relative consequences of false positives and false negatives provide a more meaningful standard for deployment, particularly in resource-constrained settings. Reframing trust around whether models improve real clinical decisions, rather than whether they achieve impressive statistical benchmarks, offers a more rigorous and globally applicable foundation for responsible AI adoption. African-led AI development should not be interpreted as scientific isolation but as a foundation for globally interoperable and reciprocal innovation and partnerships. Regional institutions such as the African Academy of Sciences, the African Union, Africa CDC, national science councils, and academic consortia have a critical role in coordinating this effort. These bodies can align investments, establish shared standards for data governance and interoperability, support workforce development, and facilitate collaboration across countries. By convening governments, researchers, biotechnology companies, and community stakeholders, these institutions can help build a coherent and sustainable ecosystem rather than fragmented, project-driven initiatives. Knowledge derived from African genomic diversity has the potential to improve variant interpretation, fine-mapping, and biological discovery across populations. However, the transferability of AI models should not be assumed. Instead, portability must be treated as an empirical question requiring validation, recalibration, and governance aligned with the target population. Ensuring global benefit while avoiding new inequities will require harmonized data standards, transparent reporting of population composition, and equitable data-sharing frameworks. Addressing these challenges requires coordinated action across multiple stakeholders. African governments, regional institutions, universities, and research consortia must lead infrastructure development, workforce training, and governance frameworks. International funders and collaborators should support these efforts through equitable partnerships that include shared leadership, data sovereignty, and long-term capacity building. Without this coordinated approach, the gap between technological potential and real-world impact will persist. Conclusion: Trust is the driver of impact AI holds transformative promise for democratizing genomic medicine, but without intentional design, African leadership, and context-grounded governance, it risks reinforcing the very inequities it seeks to solve. It must be earned through transparency, inclusion, cultural understanding, and respect for autonomy. AI will not solve Africa’s genomics challenges automatically; it can in fact amplify existing inequalities if not implemented carefully. AI should be viewed as a decision-support tool rather than a deterministic system. Its value lies in augmenting, not replacing, human expertise in genomic research and clinical care. Without local expertise, strong governance, and investment in infrastructure, AI could unintentionally deepen existing disparities, turning Africa into a data source rather than a full beneficiary of genomics innovation. African governments must invest in real-world validation studies of AI in genomic medicine alongside technical innovation. Acknowledgments This work was supported by the 10.13039/100000002National Institutes of Health R01 HG011411 (T.B.M.), the Chan Zuckerberg Initiative Foundation for Accelerate Precision Health Program to Advance Genomics Research at Meharry Medical College CZIF2022-007043 (A.G. and V.L.B.), and 10.13039/100006545NIMHD grant U54MD007593 to build research capacity at Meharry Medical College (A.G.).The funders of the study had no role in study design, data collection, data analysis, data interpretation, writing of the report, or decision to submit the report for publication. All authors read the manuscript and agreed to its content. Declaration of interests V.L.B. is affiliated with the Diaspora Human Genomics Institute, a nonprofit organization.