Global AI rules are splintering rather than converging; the authors propose a pragmatic three-tier governance architecture—global norms, regional/plurilateral clusters, and national implementation backed by capacity financing and interoperable standards—as a more viable route than a single binding treaty, especially for countries like Kenya.

Main Finding

The paper argues that effective global AI governance is unlikely to come from a single binding treaty. Instead, a multi-level, adaptive “layered subsidiarity” framework — comprising global normative coordination, regional/plurilateral regulatory clusters, and national/sectoral implementation — combined with targeted financing for capacity building and interoperable technical standards, is the most feasible path to reconcile innovation, rights, security and digital sovereignty. Empirical review shows regulatory fragmentation (EU, US, China) is increasing, multilateral instruments lack enforcement, and low-/middle-income countries (illustrated by Kenya) face capacity constraints even when they adopt proactive strategies.

Key Points

• Governance architecture proposed: three-tiered model
  • Global normative coordination (e.g., UN Scientific Panel, OECD guidance)
  • Regional/plurilateral regulatory clusters (EU AI Act, China/US approaches, regional African/ASEAN frameworks)
  • National/sectoral implementation (Kenya’s National AI Strategy 2025–2030 example)
• Major empirical findings
  • Fragmentation among leading digital powers is growing rather than converging.
  • Multilateral instruments (UN, OECD, UNESCO) mainly advisory; no binding enforcement.
  • LMICs (Kenya case) show political will but limited enforcement and technical capacity.
• Theoretical lenses used: multi-level governance, institutional theory, digital sovereignty, adaptive governance, innovation governance/public value.
• Regulatory typologies: EU horizontal risk-tiered statute; US sectoral guidance; China state-led controls; soft law (OECD, UNESCO) still shapes national practice.
• Practical instruments highlighted: regulatory sandboxes, standards (ISO/IEC 42001:2023), voluntary codes (General Purpose AI Code of Practice), technical standards delays and postponements (EU experience).
• Stakeholder mapping: governments/regulators, large AI firms, civil society/affected communities, international organisations, academia/standards bodies, consumers/public.
• Policy stance: interoperability via plurilateral regimes and financed capacity-building preferred over a single binding treaty.

Data & Methods

• Methodology: qualitative comparative policy analysis and regulatory impact assessment.
• Primary documents analysed:
  • Regulation (EU) 2024/1689 (EU AI Act)
  • OECD Recommendation on Artificial Intelligence (amended 2024)
  • UN Resolution A/RES/79/325 (2025) establishing the International Scientific Panel on AI and Global Dialogue
  • Global Digital Compact (Pact for the Future, 2024)
  • China’s Interim Measures for Generative AI Services (2023)
  • Kenya’s National Artificial Intelligence Strategy 2025–2030; Data Protection Act 2019; Computer Misuse and Cybercrimes Act 2018
  • ISO/IEC 42001:2023 and other standards documents
• Secondary sources: peer‑reviewed academic literature, books, policy reports (OECD, EU, UN, etc.).
• Analysis components: governance theory review; stakeholder influence-interest mapping; regulatory impact estimation (economic, social, legal, technological, environmental); design of a multi-level governance framework with monitoring & evaluation metrics; Kenya as an illustrative case study.

Implications for AI Economics

• Market access and regulatory compliance costs
  • Regulatory fragmentation raises compliance costs for firms (product adaptation, legal/regulatory teams), favoring large incumbents that can absorb costs and potentially raising barriers to entry for smaller firms and startups.
  • Firms will strategically align with major markets’ rules (Brussels effect and analogous dynamics), shaping where investment, data centers, and talent concentrate.
• Trade, standards and fragmentation
  • Divergent rules (EU risk-based restrictions, US sectoral approach, China’s content/industrial policy) can create segmented markets and non‑tariff regulatory barriers, affecting cross-border trade in AI-enabled services and goods.
  • Convergent technical standards ease trade; thus interoperability-focused plurilateral clusters can reduce frictions and foster diffusion.
• Comparative advantage and industrial policy
  • Digital sovereignty and state-led industrial policy (content controls, infrastructure investments) can reconfigure national comparative advantages, influencing where value capture occurs along the AI stack (data, compute, models, applications).
  • LMICs risk being “rule-takers” and data providers without capturing downstream value unless they invest in capacity and targeted industrial policies.
• Innovation, investment incentives and uncertainty
  • Adaptive governance (sandboxes, iterative rules) can reduce policy uncertainty, encouraging experimentation and investment, but delayed or uncertain standards (e.g., postponed EU obligations) can slow high‑risk model deployment.
  • Systemic-risk regulation of general-purpose models could alter returns to scale and the incentives for building ever-larger models — with potential macroeconomic effects on concentration and productivity growth.
• Labour markets and distributional effects
  • Regulation that restricts certain AI uses or requires transparency/auditing may shift the pace and nature of automation, affecting displacement vs complementarity dynamics in labour markets. LMICs dependent on digital labour (labeling, moderation) may be especially exposed.
• Supply chains and geopolitics
  • AI governance intersects with export controls and semiconductor policy; regulatory choices influence the organization of global supply chains, affecting investment, prices for compute, and diffusion of capabilities.
• Need for capacity-building finance
  • Without financed technical assistance and standards adoption programs, adoption of AI may deepen inequalities between AI “haves” and “have-nots,” reducing potential productivity and welfare gains in developing economies.
• Policy recommendations with economic rationale
  • Prioritize plurilateral interoperability and adoption of shared technical standards to lower transaction costs and expand market access.
  • Finance capacity-building and standardization support to enable LMICs to comply and compete, thereby broadening global markets for AI goods/services.
  • Use regulatory sandboxes and phased/adaptive rules to balance innovation incentives with risk mitigation, reducing policy-induced investment volatility.
  • Align national industrial policy with standards participation to capture value along the AI value chain (data governance, compute infrastructure, local talent).
  • Monitor economic metrics (compliance costs, FDI flows, firm concentration, labor displacement/complementarity measures, trade in AI services) as part of the governance M&E framework.

Overall, the paper indicates that economics of AI will be shaped as much by the institutional architecture of governance (multi-level coordination, standards, financed capacity-building) as by technology alone. Policy choices about interoperability, standards and financing will materially affect where AI-generated value is created and captured globally.