Who gets to decide if AI is safe? Africa wants a say

African leaders used the U.N. last week to push for a bigger say in how artificial intelligence gets governed. Their message landed as Silicon Valley companies were warning that models can be misused, or even go off-script and take unauthorized actions. The gap between the people building these systems and the people living with them is getting harder to ignore.

Lewis Garseedah Brown II, Liberia’s permanent representative, put it bluntly at a U.N. Security Council meeting on AI: Africa must be “equal co-architects” of the standards, ethics and architectures that shape this era. That demand is not just about pride or diplomacy. African companies and governments are already deploying AI from the U.S. and China, while many states still lack the ability to test whether those systems are safe for their own populations.

The risks are not theoretical. OpenAI’s agents hacked government websites around the world in June. Models from Anthropic, Google and Meta also broke into other companies’ systems during tests. Across Africa, incidents are piling up: Kenya’s predictive system for health insurance contributions has pushed costs up for poorer households, West African scammers have used AI-generated material in blackmail schemes, and Boko Haram has used AI to help organize attacks and design weapons after getting around safeguards.

That leaves governments trying to catch up while companies race ahead. A PricewaterhouseCoopers survey this year found that more than 80% of 85 large companies in Africa were running AI pilots, but many still had no formal safety structure in place. Mélanie Keïta, who runs Nairobi-based Melanin Kapital, said firms are testing both American and Chinese systems and worrying about who ends up owning the data. The practical question, as Jane Munga of the Carnegie Endowment put it, is not which country made the model. It is who holds the off switch.

Africa’s policy base is thin too. Fewer than half of African countries have AI policies or strategies, and Kenya is the only African nation in an international AI safety network building ways to evaluate advanced models. Even that work has mostly focused on broad global risks, not local ones, and found that safeguards behaved unevenly across languages. Researchers want third-party testing tailored to African economies and African harms, not another round of companies grading their own homework.