Prime Minister Dr. Justin Nsengiyumva speaking at an AI governance discussion at the United Nations General Assembly.
NEW YORK – When the world talks about making artificial intelligence safe, the question of who is actually in a position to test it is easily overlooked
Rwanda’s Prime Minister Dr. Justin Nsengiyumva raised that concern at an AI governance discussion at the United Nations General Assembly, insisting that much of the world is left on the outside of a conversation whose consequences will affect them.
“Countries which don’t develop frontier AI models should not simply trust that systems developed elsewhere will work safely for them given the different populations and conditions,” Nsengiyumva said.
Governments, he said, need to be able to know whether they are safe, reliable and who is accountable when it fails, before putting powerful systems to work.
Nsengiyumva said Rwanda’s National AI Agency is ready to serve as a regional host for shared AI testing, with technical expertise helping countries across Africa build the ability to independently evaluate advanced systems.
“Most states cannot assess a frontier model today, and will not be able to for years. That means inclusion without capacity is ceremonial,” he warned.
Conversations confined in two rooms
Prime Minister Dr. Justin Nsengiyumva interacting with delegates.
Nsengiyumva described a global conversation that is effectively taking place in two rooms without fully considering a much larger group outside.
In one, leading AI companies discuss standards and safety among themselves. In another, governments of countries competing over the technology negotiate with one another.
These are countries that may not build frontier models, but will still deploy them, depend on them and absorb the consequences when they fail.
That gap is becoming harder to ignore as governments look for ways to cooperate on AI safety. Countries are also entering the AI era from very different starting points.
While some are already integrating AI into public services and workplaces, others are still building the data centers, computing capacity, connectivity and skills needed to use the technology effectively.
President Alar Karis of Estonia, who also addressed the UN gathering, said the choice should not be between developing AI quickly and making it safe. For him, the two have to move together.
“There is need for common methods for evaluating AI systems and better cooperation. Otherwise serious incidents cross borders and stronger technical capacity in countries that do not have the expertise or resources to carry out such work themselves may occur,” he said.
His point adds another layer to Rwanda’s argument where the issue is no longer simply whether countries can access AI, but whether they have enough knowledge and institutional capacity to understand what they are being given.
The divide is therefore not simply between countries that have AI and those that do not. It is increasingly between countries that can turn access into useful, locally relevant applications and those that remain dependent on systems they have limited capacity to examine or adapt.
That distinction could become increasingly important as AI moves into areas where mistakes can affect people directly.
Africa does not want to remain the customer
Kenya’s Special Envoy for Technology, Philip Thigo, pushed the argument further, saying Africa cannot afford to become a continent that only consumes AI built elsewhere.
“That does not mean every country needs to build its own frontier model but to have enough technical and institutional capacity to make meaningful choices about technologies being deployed on the continent,” Thigo said.
On his part, Canada’s Ambassador to the UN, David Lametti, approached the same problem from a direction of responsibility.
“If you say safe, how do you make something that is out of control safe? Safety cannot be something developers think about after a system has already caused damage. Developers of such systems should carry responsibility when things go wrong,” Lametti noted.
In his assessment, companies which develop powerful AI systems need to build safety into testing and deployment, while governments work towards common standards and accountability.
That leaves Rwanda with a proposition that is more practical than another call for broad AI principles; give countries that do not build frontier AI the ability to test it.
It means turning Rwanda’s National AI Agency into part of the infrastructure of AI safety in Africa, helping countries acquire the expertise to evaluate systems before they become deeply embedded in hospitals, classrooms, businesses and public services.
The challenge is becoming more immediate as AI systems grow more autonomous and their failures can cross organizational and national borders.
A UN scientific panel noted that no single company or country is likely to see enough incidents on its own to identify every emerging pattern.
“Pacing the frontier is the right instinct but a pace set by a handful of companies and countries is not a pace the world has agreed to. We cannot call AI governance inclusive if only those building the technology have the capacity to judge it,” Nsengiyumva said.
That may be the more difficult conversation ahead of determining not simply how fast AI should move, but whether the countries expected to live with its consequences have the means to examine it for themselves.
