AI systems are not neutral 

Embedded in every AI model’s dataset is an assumption about who that technology is meant to serve. In the case of most globally recognisable AI models, that user is typically from the Global North. And yet, many AI models directly benefit from the knowledge inherited from African societies and the raw data supplied by their people. 

In agricultural technology and smart-farming practices, AI-generated recommendations on seasonal crop survival have roots in Cameroon, where local farmers have tilled the land for centuries, watching the skies for signs and relying on ancestral practices to keep the soil fertile. In automated medical diagnostic tools, recommended pharmaceuticals point patients to traditional medicines like Mufandichamuka, whose indigenous names were translated from Shona to English before they were packaged and sold by big, powerful brands. 

In both cases, the origins of the data are not evident, and almost certainly, the knowledge communities were not adequately compensated or credited for their contributions to the AI systems that now also threaten to displace them, whether through environmental impact or shifting labour conditions.  

Recentering African knowledge systems in AI governance and scholarship

In this context of uneven development of AI systems, epistemic traditions and knowledge hierarchies perpetuated by the Global North’s captured data economy, African AI scholars and researchers are tasked to reduce data invisibility and algorithmic discrimination. Further this agenda, Just AI project aims to develop a rights-centric policy evidence base for the socially, economically and politically just governance of AI systems in Africa. 

To strengthen the intellectual and empirical foundations of the Just AI project, four fellows were selected for the Africa Just AI Research Fellowship, a scholarly programme which leveraged the 2026 Just AI Conference as a convening site for the fellows to meet, absorb and investigate research findings put forward by a cohort of more than 50 African AI researchers, policymakers, regulators and stakeholders.

The fellows, Emmanuel Kpako Brown, Dr Elisabeth Achancho, Simbarashe Knox Kaneunyenye, and Dr Shirely Genga, synthesised the conference’s learnings and brought together their own unique research interests in AI and Digital Forensics, Human Rights, Criminology, and Discrimination in a forthcoming joint output. 

At a recent RIA Research Clinic, the fellows presented the findings from their conference brief, Five Things SADC Policymakers Should Know About Justice-Centred AI Governance, which highlights five emergent and cross-cutting themes impacting the continent. As Kpakpo Brown says, the brief’s key finding was this: “Justice in AI cannot remain aspirational. It must be operationalised through accountable institutions, context-sensitive governance, human oversight and systems designed with African realities in mind.” 

Just AI Fellows call for embedded inclusion and accountability

To institutionalise justice within AI governance, the Just AI Fellows interrogated the lived experiences of many marginalised African communities across sub-Saharan Africa and contextualised them within the region’s dominant governance frameworks. The result is five key learnings that can strengthen sustainable digital transformation in the region and the continent, through a Just AI framework. 

1. Accountability must be built into AI systems

So long as justice remains external to AI systems, it remains difficult to enforce. With most AI systems currently lacking mechanisms to trace coded faults in automated decision-making, remedy is often difficult, and costs are displaced onto affected users. For this reason, Fellows call for accountability to be built into the AI architecture itself. Legally mandated traceability mechanisms for AI models can allow users and developers alike to easily identify the causes of AI harms and the liable party. As Kpakpo Brown says, “The fellowship helped us think about accountability not as a reactive process after the harm has been caused, but as something embedded proactively into governance systems.”

2. Including African languages and knowledge systems in AI models

“African languages are not merely datasets; they are expressions of identity, culture, history, and sovereignty. The fellowship encouraged us to think about language AI not only as technical innovation but as part of a broader project of inclusion, preservation, and decolonisation,” says Kaneunyenye. Referencing work presented by Dr Tigist Shewarega Hussen at the conference, he emphasised that without African AI models that accurately speak and translate diasporic dialects such as Chichewa or Lingala, the continent faces the threat of linguistic extinction. With many African languages passed on only through oral tradition, the push to datify African languages becomes the push to preserve the people who passed them on. 

3. Governance must address power, not just ethics

Normative principles and ethical frameworks are useful in shaping a common standard for AI use and deployment. Yet, without a deeper interrogation of the power structures that underpin them, their relevance is fleeting. Achancho acknowledges that AI governance requires deliberation and negotiation over political power among varying institutions, governmental departments, and private and public sector players, before justice can be enforced. She says, “Democratic outcomes these days are no longer shaped by constitutional text or political institutions alone, but they are also shaped by systems that structure visibility, participation, verification, and accountability.”  When it comes to the use of AI in high-risk sectors such as electoral management systems, digital identity schemes, social protections and public communication, the question here is: who controls the infrastructure, and who remains accountable for its failure?

4. Policy translation requires institutional alignment

AI governance in sub-Saharan Africa is notoriously constrained by bureaucratic fragmentation, uneven regulatory capacity, and competing national priorities. While there is no doubt that various African Union (AU) policy documents and frameworks — including the AU Data Policy Framework, the Continental AI Strategy, and the African Continental Free Trade Area (AfCFTA) – offer strong foundations for development-oriented AI advancement, research and policy stakeholders repeatedly remind us that Africa’s challenge is in implementation and enforcement, not just policy development. Institutional coordination, readiness and capacity-building affect how AI’s complex technical regularities come into force. As Kpakpo Brown put it, beyond justice-oriented principles, fellows interrogated “which institutions have the authority, expertise, and political capacity to operationalise justice.” Responding to this, Kaneunyenye also said: “We need binding instruments and codified structures that are integrated with the law.”

5. Multistakeholder governance and African Futures

Enforceable AI governance requires buy-in and political will not only from regulators, policymakers, and technical experts, but also from multistakeholder groups, who act as the mouthpiece for marginalised communities. Genga emphasised that Multistakeholder collaboration in AI governance is essential to democratising AI development and regulation, moving beyond the agreement of Ubuntu-based principles, towards actual, on-the-ground, community-based decision-making. She says, when it comes to multistakeholder AI governance, “We need industry, we need companies, we need academia, we need technical experts, and most importantly, we need the affected communities.” Only through these representative approaches can we begin translating African realities into AI governance. 

Towards just futures

The Just AI Fellowship made one governance reality abundantly clear: Justice in AI is not only technical. It is institutional, political, social, and deeply connected to African futures. 

Read more about their recommendations for SADC policymakers in the forthcoming brief.