
The European Union (EU) and Kenya are converging on the same artificial intelligence (AI) governance failure from opposite directions: rules without a supervisor capable of enforcing them. The EU could not deliver its own regime on schedule, and Regulation (EU) 2026/1744 deferred the core high-risk obligations of the EU Artificial Intelligence Act by sixteen months, six days before they were due to apply. Kenya has fragmented into three partly inconsistent governance tracks that disagree about the identity of the regulator. In both cases, the binding constraint is institutional capacity rather than drafting. A second finding is that intellectual property in AI training data is now the fastest-widening gap in the Kenyan framework: it is load-bearing for the flagship ambition of locally grounded language models, and it is the gap least amenable to a purely national solution.
The paper reaches those findings by comparing Regulation (EU) 2024/1689 (the Artificial Intelligence Act) with the Kenya Artificial Intelligence Strategy 2025–2030 across scope, achievement, shortfall, deterioration over time, and reform. It argues that the two instruments are routinely compared as though they were commensurable, and that this framing distorts assessment of both. The European instrument is a binding risk allocator built on existing product-safety architecture; the Kenyan instrument is a non-binding capability builder addressed at the preconditions for an AI ecosystem that does not yet exist. Each is strongest where the other is silent.
The paper offers eight reform proposals for Kenya and six for the EU. It concludes that the transferable lesson from Brussels to Nairobi is not the substance of the risk taxonomy but the discipline of naming, funding and empowering an enforcer, and that the training-data question requires regional coordination through the African Regional Intellectual Property Organization and the African Union rather than national legislation alone.