The international conversation around artificial intelligence safety has developed a significant blind spot when it comes to the African continent, according to researchers advocating for a more inclusive approach to AI governance.

Liz Orembo, who directs international and regional engagement at Research ICT Africa and serves as a UN AI Fellow, has articulated a critical distinction that the global AI safety community has largely sidestepped. According to AI Weekly, Orembo divides AI safety concerns into two separate domains, arguing that the second category represents the actual risks confronting African populations.

Two Categories of AI Risk

The first category encompasses what Orembo calls model risks: the technical hazards that dominate discussions at frontier AI laboratories and national AI Safety Institutes. These include concerns about large language models producing toxic content, generating misinformation, or exhibiting bias in decision-making systems. While important, these risks receive disproportionate attention and funding from major AI companies and governments.

The second category, which Orembo emphasizes, encompasses the deployment harms that affect real users across Africa. This includes how AI systems amplify existing inequalities, perpetuate discrimination in financial services and hiring, enable surveillance infrastructure that threatens privacy and freedom, and concentrate economic benefits among wealthy nations while extracting value from African data and labor.

The Governance Gap

This framework exposes a fundamental problem: global AI governance architectures designed primarily by Western institutions and technology leaders naturally prioritize the risks those actors understand and can address. Technical safety becomes the proxy for comprehensive safety, leaving structural and deployment harms largely unexamined in formal policy discussions.

  • Frontier model safety focuses on preventing specific outputs or behaviors
  • Deployment harms involve systemic integration into financial, judicial, and educational systems
  • African nations lack resources to conduct independent AI audits or enforcement
  • Data flows from Africa to global AI systems remain largely unregulated

The consequences extend beyond academic debate. As AI systems integrate deeper into critical infrastructure across Africa, from banking systems to healthcare diagnostics, the absence of robust safeguards against discriminatory or extractive deployment patterns creates real vulnerabilities. African governments face pressure to adopt AI technologies quickly without adequate frameworks to protect their populations from misuse.

Reframing the Conversation

Orembo's contribution to this debate pushes back against the assumption that solving technical safety problems at the model level automatically addresses safety for all users globally. Instead, her analysis suggests that meaningful AI safety for African communities requires separate governance mechanisms focused on how systems operate in specific local contexts, how data collected from African users is used, and how benefits from AI development are distributed.

This perspective challenges current international AI initiatives to expand their mandates beyond laboratory safety toward inclusive risk assessment that reflects the diversity of global users. Without such expansion, the framework argues, AI safety remains a luxury conversation for wealthy nations while actual harms accumulate elsewhere.