Governing what you cannot inspect
The UN warns AI is outrunning its rules. Kenya has a strategy, a bill, and an election coming.
Today in Geneva, for the first time, every UN member state sits at one table to talk about governing artificial intelligence. On that table is a warning. The UN's new independent scientific panel, in its first report, finds that current safeguards cannot keep pace with what AI can now do. Secretary-General António Guterres put it plainly at the launch: "The more AI advances without shared rules, the less say governments and people will have in the outcome."
The report is a first of its kind, an independent assessment by scientists drawn from every UN region, co-chaired by Turing Award winner Yoshua Bengio and Nobel laureate Maria Ressa. It was written to give all 193 member states the same evidence to argue from. The evidence is uncomfortable for most of them.
Here is the arithmetic underneath the diplomacy. By estimates the panel cites, the United States holds 75 percent of the computing power among the world's top 500 AI supercomputers. China holds 15. Nearly every leading model comes from companies in those two countries. So Geneva is 193 states discussing a technology that two of them physically possess. For everyone else, the report has a phrase that deserves to be read slowly: most member states depend on systems they "cannot build, inspect, audit or fully adapt to local context."
Kenya has not been sitting on its hands. There is a national AI strategy, launched in March 2025. There is a High Court order from February directing the government to stop delaying on AI regulation. And there is now the Artificial Intelligence Bill, 2026, before the Senate, proposing an AI Commissioner with the power to enter premises, inspect AI systems, and keep a public register of high-risk ones. On paper, Nairobi is building exactly what Geneva says the world needs.
Some of the bill stands on solid ground. When an algorithm rejects your loan application or screens you out of a job, that decision happens here, to you, and the bill would give you the right to an explanation and a human review. That is real protection, in the space where Kenya's jurisdiction is real.
The harder part is the word inspect. The models behind the tools Kenyans use every day, in banks, in hospitals, in government offices, were built by companies that do not answer to Kenyan courts, and as Kenyan analysts reviewing the bill have noted, no regulator in Nairobi can compel them to open their code. The Commissioner will be able to examine how AI is used in Kenya. What the AI is remains out of reach. That is not a flaw unique to this bill. It is the same gap the UN panel found nearly everywhere, including in wealthy economies. But it lands harder here, because the clock is different here. The report warns that AI-made content, persuasive and cheap at scale, erodes the shared reality democratic debate depends on. Kenya votes in thirteen months. The panel calls this the evidence dilemma: by the time the proof of harm is clear, it may be too late to act on it. In Kenya, "too late" has a date.
If a law can govern how AI is used but never see inside what it is, how much protection can it honestly promise, and can Kenya finish answering before the country votes?
Preliminary Report of the UN Independent International Scientific Panel on AI, executive summary read at un.org (July 1, 2026); Secretary-General's remarks via UN Media and UN News; Global Dialogue dates per un.org. Story first seen via an Eastleigh Voice republication of UN News. Kenya AI Strategy per the ICT Ministry; Artificial Intelligence Bill, 2026 status per the Parliament of Kenya, with provisions and analysis as reported by Bowmans, Techweez, Bantu Gazette and Tech Policy Press. Compute figures are recent estimates cited by the panel.