Countries speak the same AI ethics language but govern very differently: a comparative analysis of 24 policy texts finds five distinct models of AI governance—from Europe’s rights-first approach to China’s state-centric model—driven by underlying political-economic institutions rather than convergent ethical principles.
Artificial intelligence governance and social policy divergence: a comparative political economy perspective on global AI regulation
Comparative coding of 24 AI policy documents reveals five distinct governance models—rights-based (EU), market-driven (US), state-centric (China), hybrid (Australia–Japan–Singapore) and developmental (India)—where institutional legacies, not shared ethical language, explain cross-regional differences in risk allocation, regulatory authority, labour integration and social protections.
To explain why artificial intelligence (AI) governance produces systematically divergent social policy outcomes across regions despite widespread convergence around ethical standards for AI. The study examines how political–economic institutions shape the allocation of social risks, regulatory authority, labour integration and ethics institutionalisation in AI governance, thereby driving these differences. This study uses a comparative qualitative policy analysis based on 24 key AI policy documents published between 2018 and 2025 across the European Union, United States, China, and Indo-Pacific economies. Guided by theory, the documents are systematically coded across four institutional dimensions and converted into simple indices to compare governance approaches across the regions. The findings show clear and systematic differences in how regions govern AI. Five distinct governance models emerge: rights-based (EU), market-driven (US), state-centric (China), hybrid (Australia–Japan–Singapore) and developmental (India). Although many regions use similar ethical language, substantial differences persist in risk allocation, regulatory enforcement, welfare integration and social protection. These differences reflect the historically embedded political–economic institutions shaping each regime. The paper reframes AI governance as a form of social policy shaped by political and economic institutions. It develops a new, evidence-based typology of AI governance models and shows that differences across countries are driven by institutional structures and not by ethical principles alone.
Summary
Main Finding
Although countries and regions increasingly converge on ethical language for AI, their actual governance produces systematically divergent social policy outcomes. These differences are driven not by ethics alone but by historically embedded political–economic institutions that shape how social risks are allocated, regulatory authority is organised, labour is integrated, and ethics are institutionalised. Five distinct governance models emerge: rights-based (EU), market-driven (US), state-centric (China), hybrid (Australia–Japan–Singapore) and developmental (India).
Key Points
- Convergence in ethical rhetoric masks substantive cross-regional divergence in governance practice and social outcomes.
- Four institutional dimensions structure these differences:
- Allocation of social risks (who bears economic and social costs of AI-related harms).
- Regulatory authority (public vs. private enforcement, centralisation, scope).
- Labour integration (how workers are protected, retrained, or integrated into AI-driven labor markets).
- Ethics institutionalisation (degree to which ethical norms are embedded into institutions and binding rules).
- Five governance models:
- Rights-based (EU): emphasis on rights protection and stronger regulatory instruments.
- Market-driven (US): preference for market mechanisms and lighter-touch regulation.
- State-centric (China): centralised control and state-led governance priorities.
- Hybrid (Australia–Japan–Singapore): mixed approaches combining elements of regulation, markets and targeted state intervention.
- Developmental (India): focus on growth, inclusion and developmental priorities with distinct regulatory trade-offs.
- Key policy differences across regions include variation in risk allocation, enforcement strength, welfare integration and social protection mechanisms.
- Institutional legacies, not merely ethical commitments, explain why similar ethical frameworks yield different governance outcomes.
Data & Methods
- Comparative qualitative policy analysis of 24 key AI policy documents published between 2018 and 2025.
- Regions covered: European Union, United States, China, and Indo‑Pacific economies (Australia, Japan, Singapore, India).
- Theory-guided coding: documents were systematically coded across the four institutional dimensions described above.
- Coding results were converted into simple indices to enable cross‑regional comparison of governance approaches.
- Strengths: cross-regional, theory-informed, document-based comparison covering major policy developments up to 2025; yields an evidence-based typology.
- Limitations: reliance on policy documents (may differ from practice), limited sample size (24 documents), and qualitative index construction that may omit subnational variation and implementation dynamics.
Implications for AI Economics
- Policy heterogeneity matters for economic outcomes: divergent governance models will produce different incentives for firms, affect compliance costs, shape innovation pathways, and alter distributional impacts of AI.
- Labour markets and social protection: variations in labour integration and welfare design imply differing trajectories for worker displacement, retraining uptake, unemployment risk and inequality across regions.
- Investment and comparative advantage: firms will face region-specific regulatory risks and market structures—affecting location decisions, R&D investment, and business models.
- Global coordination challenges: shared ethical language is insufficient for harmonisation; institutional differences will complicate efforts to create universally binding standards or interoperable regulations.
- Modeling and empirical work should explicitly incorporate institutional variables (e.g., regulatory centralisation, welfare regime type, enforcement modalities) when forecasting AI’s economic impacts or simulating policy interventions.
- Policy design: efforts to align AI governance across borders should focus on reconciling institutional constraints (e.g., liability regimes, welfare systems) rather than only promoting shared principles. International cooperation could target specific institutional levers (mutual recognition, baseline social protections, labor transition funds) to reduce harmful fragmentation.
Assessment
Paper Typedescriptive
Evidence Strengthlow — The study provides systematic descriptive evidence of cross-regional differences in policy texts but does not use a research design that isolates causal effects (no counterfactuals, no temporal variation exploited, no outcome data on enforcement or social impacts), so claims that institutions 'drive' differences rest primarily on theory and pattern-matching rather than strong causal identification.
Methods Rigormedium — The paper applies a transparent, theory-guided coding scheme and converts qualitative codes into indices for comparative analysis, which is appropriate for typology-building; however, the small document sample (24), potential selection and coder-subjectivity biases, lack of reported inter-coder reliability or triangulation with implementation data, and limited attention to within-country variation constrain methodological rigor.
Sample24 key AI policy documents published 2018–2025 from the European Union, United States, China, and selected Indo-Pacific economies (Australia, Japan, Singapore, India); likely includes white papers, national strategies, regulatory proposals and guidance documents rather than implementation/enforcement records.
Themesgovernance labor_markets inequality
IdentificationTheory-guided comparative qualitative coding of 24 national/regional AI policy documents into indices across four institutional dimensions; comparison of patterns across regions to infer how political-economic institutions shape governance differences (no formal causal identification, no counterfactuals or quasi-experimental design).
GeneralizabilitySmall and purposive sample of policy documents limits ability to generalize beyond selected regions and document types, Focus on formal, published policy texts; does not capture implementation, enforcement, or subnational variation, Selection bias in choosing ‘key’ documents may over-represent formal rhetoric or flagship policies and under-represent incremental or sectoral rules, Language and translation issues could affect coding comparability across jurisdictions, Temporal window (2018–2025) may miss rapid post-2025 regulatory changes or emergent practices, Findings describe governance models but do not directly generalize to economic outcomes (productivity, wages, employment) without further empirical linkage
Claims (8)
| Claim | Direction | Confidence | Outcome | Details |
|---|---|---|---|---|
| This study uses a comparative qualitative policy analysis based on 24 key AI policy documents published between 2018 and 2025 across the European Union, United States, China, and Indo-Pacific economies. Governance And Regulation | mixed | high | research design and document sample |
n=24
0.3
|
| The documents are systematically coded across four institutional dimensions and converted into simple indices to compare governance approaches across the regions. Governance And Regulation | mixed | high | coding across four institutional dimensions and index construction |
n=24
0.3
|
| The findings show clear and systematic differences in how regions govern AI. Governance And Regulation | mixed | high | degree and nature of differences in regional AI governance approaches |
n=24
0.18
|
| Five distinct governance models emerge: rights-based (EU), market-driven (US), state-centric (China), hybrid (Australia–Japan–Singapore) and developmental (India). Governance And Regulation | mixed | high | categorical classification of regional AI governance model |
n=24
0.18
|
| Although many regions use similar ethical language, substantial differences persist in risk allocation, regulatory enforcement, welfare integration and social protection. Social Protection | mixed | high | similarity of ethical language vs. divergence in (a) risk allocation, (b) regulatory enforcement, (c) welfare integration, (d) social protection |
n=24
0.18
|
| These differences reflect the historically embedded political–economic institutions shaping each regime. Governance And Regulation | mixed | medium | institutional drivers of governance differences |
n=24
0.05
|
| The paper reframes AI governance as a form of social policy shaped by political and economic institutions. Governance And Regulation | mixed | high | conceptual framing of AI governance as social policy influenced by political-economic institutions |
n=24
0.18
|
| It develops a new, evidence-based typology of AI governance models and shows that differences across countries are driven by institutional structures and not by ethical principles alone. Governance And Regulation | mixed | medium | existence of an evidence-based typology and the asserted causal role of institutional structures over ethical principles |
n=24
0.11
|