Responsible AI safeguards are lagging behind government AI adoption across 135 countries

Governments are adopting artificial intelligence strategies faster than they are building the systems needed to supervise AI, limit harmful uses and help people obtain remedies, according to the 2026 Global Index on Responsible AI. The assessment covers 135 countries and finds that the average national score for responsible AI is about 35 out of 100, despite generative AI having been used by 53% of the global population.

The index measures more than whether a country has published an AI strategy. It examines 38 indicators across policy, public participation and the conditions needed to put responsible-AI commitments into practice. Researchers reviewed 1,872 documents, with most evidence gathered between November 2023 and September 2025. The findings therefore provide a broad snapshot of governance during a period of rapid technological and policy change, but may not include measures adopted after the evidence-collection period.

A central result is the distance between formal commitments and implementation. Although responsible-AI frameworks are spreading, the researchers found evidence that they were being put into practice in only 55% of cases worldwide. In the Global South, the implementation rate was 45%. The character of those frameworks also differs: 78% of frameworks in the Global South were non-binding, compared with 42% in the Global North. Non-binding guidance can help establish expectations, but generally offers weaker guarantees than laws, enforceable regulations or independent oversight.

The index also points to specific weaknesses in public accountability. Only 18% of assessed countries require governments to disclose the AI or algorithmic systems they use. Without such information, residents may not know when automated tools influence decisions about benefits, policing, migration, employment or public services. Just 27% of countries have frameworks that address AI’s environmental effects, while 29% have measures focused on labour protections. These gaps matter because AI systems depend on energy-intensive computing and can alter how work is allocated, assessed or eliminated.

The overall picture is uneven rather than uniformly negative. Nigeria ranked 38th globally and first among African countries, rising 42 places from its earlier position in the index. The country’s result was associated with its national AI strategy, investment in digital infrastructure, skills initiatives and the 3 Million Technical Talent programme. Nigeria’s ranking illustrates that efforts to expand access to AI capacity can develop alongside attempts to establish public-interest safeguards, although a ranking alone does not show how consistently those measures work in practice.

Thailand ranked 60th, with a score of 36.99, slightly above the global average. Its comparatively stronger infrastructure and workforce readiness helped its position, while weaker policy implementation and limited civil-society participation pulled it down. The participation score reported for civil-society groups was 19.55, suggesting that the ability of communities and independent organizations to influence AI policy remains limited even where technical capacity is growing.

The Philippines ranked 68th, scoring 32.98. Together, the results from Thailand, the Philippines and Nigeria show why national averages need context. Countries can have growing digital infrastructure, skills programmes or national strategies without yet having equally strong mechanisms for transparency, public involvement, enforcement and redress. They also show that responsible-AI capacity is not confined to wealthy countries, but that progress can be fragmented within every region.

The Global Center on AI Governance, which published its report alongside the index, describes the underlying challenge as a gap between public promises and practical accountability. Effective governance requires more than principles: institutions need authority, funding, technical expertise and independence. People affected by automated decisions also need clear information, ways to challenge errors and access to remedies when systems cause harm.

Because AI products and data flows cross borders, these weaknesses have consequences beyond any one country. A system developed in one region may be deployed elsewhere, while workers, consumers and communities may face decisions made by organizations operating across jurisdictions. The index’s findings suggest that expanding AI access without comparable investment in oversight risks spreading the benefits unevenly while leaving affected people with few ways to question or correct harmful outcomes.