Version 1 · published methodology

Index 2026 methodology

Transparent rules for turning primary-source evidence into comparable country assessments.

Scope and unit of analysis

The Index covers the 193 United Nations member states, the Holy See and the State of Palestine. The European Union has a complete supranational profile but is not ranked as a sovereign state. The annual evidence cutoff is 31 July 2026.

The knowledge graph can continue to update. Once the annual edition is published, its results and frozen quantitative bounds are immutable.

Scoring anchors

ScoreDocumentary evidence anchor
0Reviewed documentary evidence of absence.
25Announcement, draft, consultation or pilot.
50Adopted but partial, not fully applicable or not operational.
75In force and operational, with implementation evidence.
100Comprehensive, independently supervised and supported by published outcomes or enforcement.

Quantitative normalization

Quantitative indicators use the median of eligible observations dated from 1 January 2024 through the cutoff. Per-capita or GDP measures are supplied where required by the indicator definition. On the first eligible recalculation, log1p is selected when skewness exceeds two, values are winsorized at the 2.5th and 97.5th percentiles, and the transform plus min-max bounds are persisted and reused.

Missing data

Missing evidence is never scored as zero. Documentary and quantitative indicators are not imputed in the 2026 implementation. Coverage below 90% is labelled provisional; coverage below 70% does not produce a country result.

A zero means reviewed evidence of absence. “Not found” means missing. No documentary or quantitative value is imputed in the 2026 implementation.

Verification and confidence

Every observation is linked to a verified Tier 1–3 source and immutable archive record with URL, publisher, source and access dates, evidence excerpt and SHA-256 hash. Every observation passes human review. Expert-rubric evidence must then be published by a different editor.

  • A: at least 90% verified indicator coverage.
  • B: 80–89% verified indicator coverage.
  • C: 70–79% verified indicator coverage.
  • Below 70%: no country result is generated.

Ranking and ties

Scores are calculated at full precision and displayed to one decimal. Equal displayed scores receive the same competition rank—1, 2, 2, 4—with alphabetical ordering used only for display. There is no hidden tie-break.

Limitations

Globally comparable AI investment data is incomplete and often commercially restricted. Announced budgets are not treated as executed spending, general R&D is not silently relabelled as AI R&D, and unlicensed private-capital datasets are excluded. This lowers coverage instead of creating a fabricated proxy.

All 45 indicators

Dimensions use the published near-equal basis-point weights. Within a dimension, available verified indicators are equally weighted and missing indicators remain missing.

01AI regulation maturity11.12%

Adoption alone does not establish maturity; the legal framework must be applicable, supervised and capable of producing remedies.

  1. REG01 · Binding legal status

    Maturity of binding AI legislation or legally enforceable AI-specific rules.

    Expert rubric
  2. REG02 · Scope and risk coverage

    Coverage of sectors, risk classes, providers, deployers and affected people.

    Expert rubric
  3. REG03 · Operational compliance rules

    Availability of implementing rules, guidance, conformity routes and compliance dates.

    Expert rubric
  4. REG04 · Competent authorities

    Independence, mandate, published resources and coordination of responsible authorities.

    Expert rubric
  5. REG05 · Enforcement evidence

    Published investigations, sanctions, judgments, corrective orders and accessible remedies.

    Expert rubric
02AI safety framework11.11%

Safety capacity must cover the system lifecycle and severe misuse, not only voluntary principles at development time.

  1. SAF01 · Risk assessment duties

    Requirements or implemented frameworks for documented AI risk and impact assessment.

    Expert rubric
  2. SAF02 · Testing and independent evaluation

    Red-teaming, audit, conformity assessment and independent evaluation capacity.

    Expert rubric
  3. SAF03 · Incident reporting and monitoring

    Operational channels for serious-incident reporting and post-deployment monitoring.

    Expert rubric
  4. SAF04 · National safety capacity

    Publicly mandated AI safety institute or equivalent scientific and regulatory capacity.

    Expert rubric
  5. SAF05 · Systemic and misuse controls

    Controls addressing systemic, cyber, biological, manipulation and dual-use risk.

    Expert rubric
03Transparency11.11%

People and oversight bodies require usable information to understand, contest and audit consequential AI systems.

  1. TRA01 · Notice to affected people

    Legal and operational notice when people interact with or are materially affected by AI.

    Expert rubric
  2. TRA02 · Public-sector AI register

    Completeness, accessibility and update discipline of public AI inventories.

    Expert rubric
  3. TRA03 · Documentation and provenance

    Requirements for model/system documentation, data provenance and meaningful explanations.

    Expert rubric
  4. TRA04 · Procurement and audit disclosure

    Publication of AI procurement, impact assessments, audits and performance evidence.

    Expert rubric
  5. TRA05 · Public participation and open data

    Access to consultations, regulatory evidence, reusable data and feedback outcomes.

    Expert rubric
04Public AI strategy11.11%

A strategy earns credit for executable public commitments, not for ambition or publication alone.

  1. STR01 · Adopted current strategy

    Official adoption, currency and public availability of a national AI strategy or equivalent.

    Expert rubric
  2. STR02 · Institutional accountability

    Named lead institutions, assigned responsibilities and cross-government coordination.

    Expert rubric
  3. STR03 · Funding and delivery calendar

    Approved funding, implementation plan, milestones and named delivery owners.

    Expert rubric
  4. STR04 · Metrics and public reporting

    Published targets, KPIs, progress reports and independent evaluation.

    Expert rubric
  5. STR05 · Coverage and consultation

    Breadth across rights, safety, skills, research, economy and inclusion, informed by consultation.

    Expert rubric
05Government adoption11.11%

Digital-government capacity is useful only when AI deployments are inventoried, governed and evaluated in their real context.

  1. GOV01 · Verified public deployments

    Breadth of operational AI deployments across public functions with official evidence.

    Expert rubric
  2. GOV02 · Portfolio governance

    Central inventory, ownership, risk classification and lifecycle oversight of government AI.

    Expert rubric
  3. GOV03 · Skills and procurement controls

    Public-sector competence, responsible procurement clauses and supplier accountability.

    Expert rubric
  4. GOV04 · Impact and outcome evaluation

    Published impact assessments, service outcomes, failure evidence and corrective actions.

    Expert rubric
  5. GOV05 · Digital execution capacity

    World Bank GovTech and UN e-government evidence relevant to implementation capacity.

    Minimum–maximum normalization
06Privacy protection11.11%

AI governance depends on enforceable controls over personal data and automated decisions, including remedies for affected people.

  1. PRI01 · Comprehensive privacy law

    Enacted and applicable comprehensive data-protection legislation.

    Expert rubric
  2. PRI02 · Independent supervisory authority

    Legal independence, mandate, resources and public activity of the data-protection authority.

    Expert rubric
  3. PRI03 · Automated-decision safeguards

    Rights and duties for profiling, automated decisions, lawful basis, explanation and contestability.

    Expert rubric
  4. PRI04 · Enforcement and remedies

    Published enforcement, sanctions, court remedies and accessible complaint procedures.

    Expert rubric
  5. PRI05 · Sensitive data and transfers

    Protections for sensitive data, breach notification, security and international transfers.

    Expert rubric
07Research capability11.11%

Governance quality requires domestic capacity to evaluate systems and participate in the development of evidence and standards.

  1. RES01 · AI publications per capita

    Three-year median of verified AI research output per million inhabitants.

    Minimum–maximum normalization
  2. RES02 · AI research specialization

    Share of national scholarly output classified as AI research.

    Minimum–maximum normalization
  3. RES03 · Normalized research impact

    Field- and year-normalized impact of national AI research output.

    Minimum–maximum normalization
  4. RES04 · R&D human capital

    Researchers in research and development per million inhabitants.

    Minimum–maximum normalization
  5. RES05 · Public AI research infrastructure

    Verified national programs, compute, datasets, institutes and sustained public support.

    Expert rubric
08AI investment11.11%

Investment is normalized to economic size and credited only where scope, execution and provenance can be verified.

  1. INV01 · Executed public AI spending

    Verified executed public AI expenditure as a share of GDP.

    Minimum–maximum normalization
  2. INV02 · AI research and development

    Transparent estimate of public and private AI R&D as a share of GDP.

    Minimum–maximum normalization
  3. INV03 · Private AI capital

    Licensed and attributable AI venture and growth investment as a share of GDP.

    Minimum–maximum normalization
  4. INV04 · Compute and data infrastructure

    Verified investment in public-interest compute, connectivity and governed data infrastructure.

    Minimum–maximum normalization
  5. INV05 · Continuity and budget transparency

    Multi-year continuity, execution reporting and auditability of AI investment commitments.

    Expert rubric
09Responsible AI policies11.11%

Responsible AI is measured through implemented safeguards and remedies rather than general ethical commitments.

  1. RAI01 · Rights and non-discrimination

    Applicable safeguards for equality, discrimination, fundamental rights and vulnerable groups.

    Expert rubric
  2. RAI02 · Human oversight and contestability

    Meaningful human authority, accessible challenge routes and corrective powers.

    Expert rubric
  3. RAI03 · Rights and ethical impact assessment

    Operational human-rights, ethical or algorithmic impact-assessment requirements.

    Expert rubric
  4. RAI04 · Inclusion and sustainability

    Implemented measures for accessibility, gender, language, participation and environmental impact.

    Expert rubric
  5. RAI05 · Audits and demonstrated remedies

    Published audits, remedy outcomes, system changes and redress produced by responsible-AI controls.

    Expert rubric