Form Before Capability? AI Governance Adoption, Functional Capability, and the Limits of Index‐Based Evidence in the Global South

ABSTRACT

Governments across the developing world have adopted artificial intelligence governance instruments at high speed, though the relationship between formal adoption and the capability to operate those instruments remains poorly understood. This article asks whether the expansion of AI governance architecture across the Global South has been matched by the functional capabilities on which regulatory effectiveness depends, and what explains the answer. Drawing on panel data for 82 countries across five Global South regions and six waves of the Government AI Readiness Index, it reports that within the period of stable index architecture (2020–2024) measures of regulatory form rose by 20.3 points on a 0–100 scale while measures of functional capability rose by 0.4 points, a within‐country differential present in 90% of sample countries and surviving removal of the single step‐scored indicator of national strategy adoption. The pattern of the differential discriminates among explanations: formal adoption rose four times faster in states in the top quartile of administrative capacity than in the bottom quartile, where capability fell; the most aid‐dependent states adopted least; and the differential is unrelated to regime type. This is inconsistent with the view that legislation is simply quicker to produce than infrastructure and with the view that adoption is decoupled from performance for external audiences, and consistent with adoption conditioned on administrative capacity, in which capability in the weakest states erodes while capability in stronger states grows more slowly than the instruments that depend on it. Vietnam and India are developed as confirming and deviant cases. Cross‐national dispersion widened on both dimensions, and patterns that appear only when the 2025 wave is included cannot be separated from the redesign of the index in that year.