The development of artificial intelligence (AI) requires considerable resources. Estimates indicate that if current trends continue, the largest trainings for advanced AI models could cost over $1,000 million by 2027. Despite limited resources, African countries are finding ways to innovate and contribute to the global AI landscape. Although Africa does not currently lead global AI development, it cannot ignore AI security due to its vulnerability to AI-related damages, especially as AI capabilities grow.
Different forms of open-access AI have been suggested as a practical means for the majority world contexts to participate in AI security. Open-access AI is defined as any form of model sharing, including staged releases, cloud-based access, access through APIs, and models with widely available weights. Research supports that open-access AI fosters a more inclusive approach to defining acceptable model behavior. In particular, it advances security research by allowing model scrutiny by external evaluators and modifications such as security fine-tuning.
Open-access AI is gaining traction on the continent. For example, the open-source coding community in Sub-Saharan Africa is growing, with countries like Rwanda and Nigeria experiencing an increase of over 45% in developers between 2022 and 2023. However, the progress of an expanding developer community, which also includes AI security researchers, could be hindered by the limitations of open-access AI in Africa. This essay examines the limitations of open-access AI as an approach to AI security in Africa. It explains that AI security research leveraging open-access AI may face obstacles due to dependency dynamics between model sharers and African security developers, as well as systematic development challenges in Africa. It then recommends how African security researchers can strategically maximize the potential of open-access AI.
Dependency Dynamics
Open-access AI thrives when it is the result of external incentives rather than voluntary decision-making. Contrary to this, AI developers unilaterally control which AI components to share and who can access them. A challenge in AI security research in Africa is the skepticism in requesting access to models. Moreover, developers make the general decision on whether to maintain proprietary control over their models. For example, although OpenAI was founded on principles of transparency, it has backtracked on its stance on openness. In 2024, the French company Mistral, a strong advocate of open-access AI in Europe, followed the same path. More concerning still, major AI powers are tending towards nationalist AI policies to gain AI supremacy. Unfortunately, developing countries are the ones who stand to lose the most, as nationalist strategies cut them off from relevant AI resources. Additionally, given the increasingly important role of AI in national security, governments are likely to become more cautious about sharing proprietary model information. The Department of Commerce of the United States, Bureau of Industry and Security (BIS), “Framework for the Dissemination of Artificial Intelligence” is a testament to this. The rule introduces a global licensing requirement for model weights trained with more than 1,026 FLOPs. In the interest of U.S. national security, the regulation restricts the export, re-export, or transfer within the country of model weights. However, it creates an exception for 18 key U.S. allies, none of which are African countries, which are considered low-risk destinations. Although the new export regulations only target advanced models, they signal stricter control over model sharing as AI advances. Additionally, African governments may face challenges in meeting the necessary licensing requirements, which could limit future access to U.S. model weights. For example, Africa currently lacks the capacity to enforce safeguards against risks associated with advanced AI, such as chemical, biological, radiological, and nuclear threats, which are an area of concern in the BIS rule. China’s extensive involvement in AI development in Africa also raises questions about Africa’s future eligibility for licenses. The BIS regulation emphasizes the supply chain independence of Third Category countries, which include China, for their AI partners. However, China is the largest exporter of AI technologies in Africa, with most African countries benefiting from agreements such as the Belt and Road Initiative (BRI) deals. Highly influential Chinese companies, such as Alibaba and Huawei, are expanding their presence on the continent, with Huawei announcing in 2023 its plans to invest $430 million in AI infrastructure on the continent over the next five years. Some researchers fear that such investments could allow China to exert influence and limit African collaboration with Western partners. Therefore, the dependency of African AI security developers on open-access AI deepens power imbalances, leaving African developers at the mercy of model producers and the countries in which they are located. This not only compromises the autonomy of African countries but also undermines the ability of developers to contribute significantly to AI security research.
The Limits of Open-Access AI in Africa
Open-access AI is generally credited with reducing development costs. DeepSeek-R1, for example, rivals frontier proprietary models at a fraction of the inference costs (approximately only 3%). However, the resources needed to utilize open-access AI are still out of reach in African contexts. Access to critical AI infrastructure, such as graphics processing units (GPUs) and cloud computing services, remains a crucial challenge. A study conducted with data scientists from Zindi Africa revealed that only 1% of them have ‘on-premises’ access to GPUs. While 4% pay for cloud access to GPUs, they can only afford approximately $1,000 per month, which translates to approximately two hours of daily use for an older Nvidia A100 GPU. The remaining 95% rely on standard laptops without GPUs or access GPUs through free cloud tools, which impose cumbersome usage restrictions. Although GPU access is a global problem, the situation in Africa is critical. To contextualize, a report highlighted the disproportionately high cost of GPUs relative to per capita GDP in different African countries. For example, the cost of an Nvidia A100 GPU was 22% of the per capita GDP in South Africa, 75% in Kenya, and 69% in Senegal. Additionally, Africa’s access to fundamental public services for AI, such as energy and digital infrastructure, remains highly limited. Africa represents only 6% of global energy consumption. By the end of 2022, only 51.5% of the population in Sub-Saharan Africa had access to electricity. Additionally, Sub-Saharan African countries experience an average of 87 power outages annually, compared to North America, which averages 1.34. This is closely related to internet access, which varies in Africa, with low-income countries like South Sudan, Burundi, and the Central African Republic having penetration below 13%. However, optimistically, the push for 5G is growing on the continent. Sub-Saharan Africa, for example, is expected to have 226 million 5G connections by 2030, a 17% adoption rate. Additionally, African developers risk being excluded from global security research networks where AI knowledge advancements are collaboratively exchanged, due to factors such as financial constraints and strict visa requirements. The lack of global attention to AI security research exacerbates the problem. AI security research constituted a meager 2% of the overall AI research landscape between 2017 and 2022. In Africa, AI funding appears to be primarily directed towards finding AI solutions for development challenges such as healthcare, agriculture, and education. Continental and national policies suggest that this trend is likely to persist.
In summary, AI security developers in Africa seeking to leverage open-access AI face a series of deeply rooted resource constraints. At the same time, despite the considerable AI investment directed at the continent, the outlook for AI security developers in Africa remains bleak due to the lack of interest in AI security within the region.
Framing AI Security Problems
Theory of framing suggests that problems can be interpreted from different perspectives and have implications for multiple values. Linking a problem to people’s values fosters a sense of ownership, especially when people perceive possible interests. In Africa, leveraging AI for inclusive development is highly valued, given its socio-economic conditions and history of subjugation by dominant economies. To address resource scarcity, AI security researchers in Africa could secure AI funding allocated for development challenges, emphasizing the risks that AI security concerns pose to these solutions. African developers could, for example, advocate for collaboration with AI stakeholders offering educational solutions by highlighting the detrimental effects that disinformation and misinformation capabilities could have on system performance, student learning experiences, and ultimately, organizational reputation. This would require more research on effective framing.
Collaboration in AI Security Research in Africa
Additionally, African developers should establish AI security research networks. Networks like the European Network for AI Security exemplify the power of collaboration. Similarly, in Africa, consortium-based collaboration has been proposed as a means to pool resources to build general AI capacity. Initiatives like the Artificial Intelligence for Development (AI4D) program have been instrumental in supporting AI researchers, innovators, and policymakers in Africa, providing crucial funding, resources, and collaboration opportunities to drive AI research. Such initiatives could serve as a means to coordinate security efforts, for example, through distributed machine learning. This approach could allow African researchers to scale their algorithms for large datasets, share computational resources, save time, and minimize redundancy. Additionally, the African Union (AU) could play a pivotal role in fostering AI security collaboration. Some researchers propose that the AU and its member states establish open computation access. Additionally, the AU could further support AI security by promoting shared research initiatives among its member states.
Developing Africa-Specific AI Security
AI security researchers in Africa could also identify “unmet needs” in model evaluation and develop specialized expertise around them to improve the chances of gaining access to models. In this way, AI security researchers in Africa could not only address model evaluation gaps but also position themselves as key contributors to the global AI ecosystem, making it more likely that external stakeholders will provide model access in exchange for valuable insights and expertise. For example, Anthropic hires crowdsourcing workers to perform red teaming and adversarial probing on their models. However, one of the drawbacks of this is that evaluations can be inconsistent due to variable characteristics in human evaluators. Based on research networks, African researchers could establish organizations with standardized policies for model evaluations to mitigate this problem.
Additionally, African countries and developers possess distinctive attributes that could make them particularly suitable for specific tasks. For example, Africa’s rich cultural, linguistic, and demographic diversity could make it ideal for robustness testing. AI security researchers in Africa could be fundamental in designing “stress tests” that simulate African scenarios for AI models, which would be useful for building more resilient models. Another opportunity arises in multilingual model testing. Despite the increase in multilingual models, most testing is done in English. The lack of testing in other languages could lead to AI using non-English languages in its dangerous capabilities such as disinformation and misinformation. Therefore, it is necessary for model testing to also be done in non-English languages. Establishing African evaluator teams that reflect Africa’s linguistic diversity to evaluate multilingual models could alleviate this concern and serve as another point of expertise that could incentivize model producers to share model access with AI security researchers in Africa.
In summary, while AI security researchers in Africa may adopt small interpersonal interventions to circumvent the obstacles of open-access AI, open-access AI cannot guarantee that African countries will participate significantly in AI security governance. Comprehensive systemic interventions, such as global commitments to AI benefit-sharing, are required. African countries could also consider negotiating model access in multilateral AI agreements with leading partners. Experts predict that AI will cause a seismic shift unlike any other historically revolutionary technology. Given the socio-economic power that AI exerts, it is crucial that the power to govern it is distributed fairly.
