A traceable scoring framework reveals wide variation in national AI policy readiness on paper: some jurisdictions set clear legal duties and rights, while others rely on voluntary guidance and coordination — the method systematically compares portfolios up to Aug 15, 2026 but does not measure enforcement or social outcomes.

Main Finding

The paper develops a replicable, transparent framework to evaluate the documented design and implementation readiness of national AI policy portfolios. Rather than focusing on one headline law or strategy, it evaluates a versioned portfolio of official instruments (statutes, regulations, guidance, funding, institutional mandates) across eight focal jurisdictions (China, India, Japan, Singapore, South Korea, the United Kingdom, the United States) and the European Union as a supranational comparator. The framework scores whether a portfolio: (1) governs serious AI risks, (2) assigns responsibilities to institutions that can plausibly implement commitments, and (3) provides lifecycle coverage and protections for people affected by AI. Scores are tied to specific provisions and documented sources; the protocol records included instruments, missing evidence and a cutoff date. The evaluation reports what the policy portfolio provides on paper — not enforcement success or social outcomes.

Key Points

• Portfolio approach: AI governance is assessed as a collection of instruments (binding and nonbinding) because different instruments create different duties and implementation resource needs.
• Versioning and traceability: The framework evaluates a time-stamped (versioned) portfolio and records the evidence and the specific provisions used to justify each score.
• Scoring design: Sub-criteria use ordered anchors and are explicitly tied to provisions; criteria cover risk governance, institutional readiness, lifecycle coverage, and remedies/rights for affected people. Public benefit and responsible-innovation objectives are evaluated separately.
• Contextual separation: National compute, economic, political and talent contexts are documented as explanatory background but are not combined into the policy score.
• Jurisdictions: Eight national jurisdictions plus the EU were compared; policy status claims are current through August 15, 2026.
• Practical focus: The framework evaluates policy text and commitments (clarity and documented readiness) — explicitly not compliance, enforcement, or downstream social outcomes.
• Limitations acknowledged: purposive selection of jurisdictions (not globally representative), dependence on available documented instruments, and absence of enforcement/outcome measurement.

Data & Methods

• Unit of analysis: a "versioned national policy portfolio" comprising laws, regulations, guidance, procurement rules, funding programs, standards work, sandboxes, institutional mandates and other official instruments.
• Evidence collection: systematic search and inventory of official instruments; the protocol documents sources searched, instruments included, missing evidence and the cutoff date (Aug 15, 2026).
• Scoring procedure:
  • Each sub-criterion has ordered anchors (i.e., defined score levels).
  • Each score is justified by citing the specific provision(s) that support the judgment.
  • Core criteria: coverage of serious risks; clarity of duties and legal force; institutional mandates/resources to implement; lifecycle coverage (from data collection to decommissioning); protections available for people affected (explainability, contestability, redress); separate axis for public benefit / responsible innovation measures.
• Contextual data (collected but not scored into policy readiness):
  • Compute infrastructure: public supercomputers, private clusters, cloud region presence.
  • Economic and political environment: GDP, trade/export controls, legal/regulatory baselines, resource availability (power, semiconductors).
  • Demand and talent: sectors demanding AI, research institutions, workforce capacity.
• Jurisdiction selection and comparison: purposive sample to span technical capacity and regulatory styles; detailed instrument-by-instrument comparisons are in appendices (policy portfolio inventories and contextual tables).
• Output: A traceable comparison of on-paper policy content that preserves differences in instrument type and legal force rather than collapsing them into a single undifferentiated index.

Implications for AI Economics

• Interpreting regulatory costs and adoption barriers:
  • The framework helps economists quantify on-paper regulatory requirements that will affect firms’ compliance costs (e.g., documentation, post-market monitoring, human oversight) and thus the marginal cost of AI deployment across jurisdictions.
  • Distinguishing binding obligations from voluntary guidance is essential for modelling expected compliance behaviour and investment risk.
• Comparative investment and firm location analysis:
  • The documented readiness of policy portfolios — when combined with contextual compute and talent data — can inform models of where firms will locate research, data centers, and production, and where foreign direct investment is more or less likely.
  • Policymakers’ stated procurement and funding commitments (captured in the portfolio) are signals that can shift private-sector incentives toward public-good applications.
• Innovation incentives and regulatory design:
  • By separating public-benefit and responsible-innovation measures from protective rules, researchers can study trade-offs between promotion of development and safeguards that might slow diffusion or raise costs.
  • The framework provides a structured way to analyze how different mixes (statutory bans, high-risk duties, voluntary codes, sandboxes) affect innovation trajectories and potential regulatory arbitrage.
• Modelling policy credibility and enforcement risk:
  • Although the paper does not measure enforcement, the explicit recording of institutional mandates and resource commitments enables economists to build conditional models linking on-paper readiness to likely enforcement capacity (as a next research step).
  • This can improve forecasting of regulatory stringency realized in practice and pricing of regulatory uncertainty.
• Public-sector adoption and welfare analysis:
  • The lifecycle-coverage and rights/protections dimensions are directly relevant to welfare calculations where AI affects access to services, labour markets, or distributional outcomes; they provide inputs to estimate potential benefits and harms of public-sector AI deployments.
• Research opportunities suggested by the framework:
  • Empirically linking on-paper readiness scores to enforcement intensity and social outcomes (error rates, discrimination incidents, redress uptake).
  • Estimating compliance cost distributions across firm sizes and sectors using the provable obligations in portfolios.
  • Incorporating institutional capacity measures (staffing, budgets) as moderators in diffusion and productivity models.
  • Using the versioned portfolio approach to study temporal dynamics: how policy changes affect investment, innovation rates and international competitiveness over time.
• Policy relevance:
  • For policymakers, the framework clarifies which commitments actually create enforceable duties vs. advisory norms — crucial when designing incentives, procurement rules, or harmonization efforts.
  • For international coordination, the method helps identify policy gaps and overlaps that drive regulatory fragmentation and potential arbitrage, informing negotiation priorities for trade and standard-setting.

Short caveat: the framework reports what policy portfolios promise on paper and ties judgments to explicit provisions; it does not substitute for empirical evaluation of enforcement, compliance behaviour or social outcomes — those remain necessary follow-up analyses for economic impact assessments.