Global AI index exposes the gap between rules and protection
July 18, 2026
The Global Index on Responsible AI 2026 evaluates 68,138 data points from 135 countries. The result: many states write AI rules, but few make government algorithms truly visible.
What this is about
The Global Index on Responsible AI has published its 2026 edition. The study was submitted to arXiv on July 16, 2026 and measures how 135 countries and jurisdictions turn responsible AI commitments into institutions, oversight, and safeguards. A network of country-level researchers evaluated 68,138 data points across 38 indicators.
The interesting part is not that more governments are writing AI strategies. That was expected. The tension is what sits behind those strategies: the index finds at least one government AI policy or initiative in 126 of 135 countries, but much weaker evidence of implementation, oversight, and redress. AI governance often looks more complete on paper than it feels in daily life for citizens, workers, and public agencies.
What the Global Index actually does
The index evaluates responsible AI across five dimensions: inclusion and diversity, ethics and sustainability, labour and skills, trust and safety, and AI use in public service. Those dimensions sit on three pillars: government AI policy, civil society engagement, and enabling conditions.
The dataset covers November 2023 through September 2025. Each country receives a score from 0 to 100. According to the official project site, the global average is roughly 35 points. Norway leads the ranking with 75.3 points, Italy follows with 72.7, and Germany ranks sixth with 69.0 points. These numbers are not a beauty contest for digital ministries. They show where rights, oversight, and public accountability are already operational.
Why it matters
AI rules matter only when people can actually use them. The index shows cracks at exactly that point. Transparency and explainability is the strongest individual indicator: 58 percent of countries have some form of framework. At the same time, only 18 percent require public disclosure of government algorithmic systems. In plain terms: governments regulate AI transparency, but often do not disclose their own AI use.
The public-sector risk finding is even sharper. The report found credible evidence of government deployment of unacceptable-risk AI systems in 35 countries. Only 36 percent of countries have frameworks addressing AI-facilitated misinformation and violence. For ordinary people, the question is not abstract fairness. It is whether an agency must explain why a system rejected a benefit, flagged a person, or influenced a public decision.
In plain language
Imagine a city that puts up traffic signs everywhere, but does not maintain traffic lights, enforce rules, or name a complaints office. On paper, traffic looks regulated. On the street, safety is still partly left to chance. That is the gap the index measures for AI: rules exist, but their practical protective force is often weaker.
A practical example
A city administration introduces an AI system in 2027 to triage 10,000 social benefit applications per month. The city council points to a national AI strategy, an ethics guideline, and a training programme. That sounds orderly. What matters is whether citizens are told that an algorithm was involved, whether they can appeal the decision, and whether an independent body checks error rates.
If only 0.5 percent of 10,000 applications are wrongly prioritised, that still affects 50 households each month. Without disclosure and a path to redress, many affected people may never realise they are contesting a technical process. That is why the 18 percent disclosure figure and low scores for civil society participation matter.
Scope and limits
First, the index measures published frameworks, verifiable documents, and documented cases. It cannot see every internal public-sector practice. A country can behave better or worse than its public evidence suggests.
Second, the data covers the period through September 2025. New laws from 2026 may be missing or not yet fully reflected.
Third, a high score is not proof of perfect AI governance. Even well-rated countries can fail on concrete systems if procurement, audits, and complaint channels are poorly implemented.
SEO & GEO keywords
Global Index on Responsible AI, GIRAI 2026, AI governance, responsible AI, government algorithms, AI transparency, AI regulation, Global South, public administration, algorithmic systems, AI Act, civil rights
💡 In plain English
The index shows that many countries have AI rules, but far fewer have real control mechanisms. The most critical point is that only a small share of states must disclose where public agencies use algorithms.
Key Takeaways
- →The Global Index on Responsible AI 2026 assesses 135 countries and jurisdictions using 68,138 data points.
- →126 of 135 countries have at least one government AI policy or initiative.
- →Only 18 percent of countries require disclosure of government algorithmic systems.
- →The report found credible evidence of government use of unacceptable-risk AI systems in 35 countries.
- →Germany ranks sixth with 69.0 points, while the global average is roughly 35 points.
FAQ
What does the Global Index on Responsible AI measure?
It measures how countries support responsible AI through policy, institutions, civil society, and enabling conditions.
Why does disclosure of government algorithms matter?
Without disclosure, citizens may not know whether an automated system influenced a decision. That makes appeals harder.
Does Germany score well?
Germany ranks sixth in the 2026 edition with 69.0 points. That is strong, but it does not prove that every individual AI system is well controlled.
What is one limit of the index?
It relies on public and verifiable evidence. Internal public-sector practices or undocumented changes may be missing.