AI Advancing Faster Than Governments Can Protect People

A new study from the Global Center on AI Governance warns that the AI race is leaving many regions behind, even as artificial intelligence becomes embedded in public services, workplaces, education, healthcare, policing, finance and everyday life.

By “leaving many regions behind,” the report refers to a widening gap between countries that are turning responsible AI commitments into enforceable rules and those still relying mainly on non-binding principles, strategies and voluntary frameworks. That divide matters because, without institutions, oversight bodies, transparency requirements and redress mechanisms, many governments remain poorly equipped to protect people as AI systems shape access to public services, jobs, education, healthcare, policing, finance and other areas of everyday life.

In this scenario, the second edition of the Global Index on Responsible AI (GIRAI) finds that responsible AI governance is expanding, but unevenly and too slowly for the pace of AI development and diffusion. While more countries are adopting AI strategies, laws and policy commitments, many still lack the institutional capacity, enforcement tools, public accountability and transparency mechanisms required to protect human rights as AI systems spread.

Built by the Global Centre on AI Governance with a network of 135 regional experts, GIRAI compares national approaches across 135 countries and five human-rights-linked dimensions: AI Use in Public Service Delivery, Ethics and Sustainability, Inclusion and Diversity, Labour and Skills, and Trust and Safety. The data used for the report covers the period 1 Nov 2023 – 30 Sept 2025.

The Index exposes a widening AI governance divide between countries with enforceable rules and countries relying mainly on non-binding principles, voluntary frameworks and early-stage capacity-building efforts.

The findings are especially relevant as governments move to regulate private AI while often failing to disclose, monitor or oversee their own use of algorithmic systems. The Index finds that the weakest-performing area is AI use in public service delivery, where automated systems can affect access to welfare, healthcare, education, housing, policing, migration and other essential services.

“Responsible AI cannot be secured through principles alone. The second edition of GIRAI shows a persistent gap between responsible AI as a commitment and responsible AI as a capability,” said Rachel Adams, Founder and CEO at the Global Center on AI Governance. “As AI becomes a structural force in public life, governments need enforceable obligations, independent oversight, public disclosure, monitoring systems and accessible routes for redress.”

A Look At The Key Findings

The report identifies a global responsible AI landscape that is more active than before, but still fragmented, under-enforced and insufficiently grounded in public accountability. Among the main findings:

AI is accelerating faster than governments can govern it in the public interest: Diffusion of AI is expanding, with 53 per cent of the global population having used generative AI tools. Yet average GIRAI scores remain low, at roughly 35 out of 100, and evidence of implementation exists in only 55 per cent of cases where frameworks are active, falling to 45 per cent in Global South countries.

Responsible AI governance is expanding in Global South countries, but binding protections remain scarce: In Global North countries, the number rose from 8.2 to 11.1, a 35 per cent increase. Global South countries account for 203 of the 306 new country cases of indicators covered by frameworks identified since the 1st Edition. Despite this progress, most of the growth is in soft law: 78 per cent of responsible AI framework cases in these countries are non-binding, compared with 42 per cent in Global North countries.

AI safety is being governed as a technical problem, while human harms remain under-addressed: AI safety and security is one of the fastest-growing areas of governance, but much of it focuses on technical safeguards. Meanwhile, the Index found credible evidence of government misuse of AI in 35 of 135 countries, and only 49 countries (36 per cent) have frameworks addressing AI-facilitated misinformation and violence.

Governments are regulating AI transparency but not disclosing their own use of AI: Transparency and Explainability is the strongest-performing indicator, with 58 per cent of countries having some form of framework. Yet implementation lags behind the existence of frameworks. For government use of AI, Public Disclosure of Government Algorithmic Systems is the weakest-performing indicator, with only 18 per cent of countries requiring disclosure of government AI systems.

Gender is increasingly recognised in AI governance, but protection from gendered harms remains weak: Gender equality is gaining visibility, with 29 new countries addressing gender and AI since the 1st Edition, but only 24 of 55 countries with gender-related frameworks show evidence of implementation. Protection from gendered AI harms remains limited.

Future generations are being prepared for the AI economy but not protected from AI-related harms: AI Literacy is one of the strongest-performing indicators, with 71 countries (53 per cent) having some framework in place and 106 countries showing evidence of some activity in this area. By contrast, only 55 countries (41 per cent) have frameworks addressing Children’s Rights in AI, and only 27 of them show evidence of implementation.

AI’s environmental footprint remains a blind spot in responsible AI governance: Only 27 per cent of countries have frameworks addressing AI’s environmental effects, and 83 per cent of those frameworks are non-binding. Very few governments require disclosure of AI’s energy use, water use, or environmental impact, contributing to making the environmental impact of AI a global blind spot.

Governments recognise the need for local-language AI but do not require developers to deliver it: Governments are investing in local-language technologies and cultural inclusion, with 52 countries (39 per cent) showing government-led initiatives. Only 47 countries (35 per cent) have frameworks addressing Cultural and Linguistic Diversity, and few require developers to use diverse datasets or adapt systems to local contexts.

Governments are investing in AI skills but neglecting workers’ rights: Labour protection frameworks exist in only 39 countries (29 per cent), compared with 72 countries (53 per cent) with frameworks on reskilling and upskilling. Few countries address workers’ rights to organise and collectively bargain in response to AI-driven workplace change.

Global AI governance is fragmenting before a shared floor of protection has been established: Average GIRAI scores range from 55 in Global North countries to 27 in Global South countries. Available evidence shows that 164 of 215 recent AI-related frameworks are non-binding, and multi-stakeholder consultations appear only 31 times in the global implementation record. Only 73 of 135 countries (54 per cent) have adopted a national AI policy or equivalent framework, and just 36 countries (27 per cent) have operational mechanisms for participation of civil society organisations (CSOs) in AI governance. Without a shared rights-based floor, interoperability risks serving markets before it protects people.

What GIRAI Measures

The Global Index on Responsible AI is a research and advocacy initiative designed to measure countries’ commitments, capacities and progress toward rights-respecting responsible AI. It is developed by the Global Centre on AI Governance, a South Africa-based think tank that works as a global hub for research and evidence-led action on inclusive and equitable AI governance.

This work was carried out with the aid of a grant from the International Development Research Centre, Ottawa, Canada and the Foreign and Commonwealth Development Office under the AI4Development funding programme. Some further funds were used for the project from the Government of Canada, and from the International Development Bank of Latin America.

The Global Centre’s mission is to reduce global inequalities exacerbated by AI and to help build a world where AI technologies and governance reflect all of humanity. A central part of its work is bringing voices that are often marginalised into global AI governance debates, especially from regions and communities underrepresented in policy conversations.

GIRAI is intended to move the global AI governance debate from aspiration to evidence. It ranks countries but it also helps policymakers identify where stronger laws, institutions, resources and implementation mechanisms are needed; helps civil society actors see where governments have made commitments and where they can be held to account; and helps international organisations, funders and researchers understand where capacity gaps are most urgent.

Growing Momentum, Challenging Implementation

Africa continues to expand its responsible AI governance landscape, but the Global Index on Responsible AI finds that implementation and enforceability remain significant barriers. The region records the world’s lowest average GIRAI score (22 out of 100), 13 points below the global average, reflecting persistent gaps between policy ambition and practical action.

Out of the 39 African countries surveyed, only 6 score above the global average: Nigeria, Egypt, Kenya, Ghana, Benin and Morocco, with North Africa emerging as the continent’s strongest-performing subregion overall.

The continent’s biggest gap lies between commitments and action. Key topics related to the Ethics and Sustainability of AI have the widest policy coverage, yet only 20.45 per cent of the existing policy frameworks are implemented. In contrast, Policy frameworks addressing Labour and Skills challenges in the context of AI record the strongest implementation rate at 65.52 per cent, making it the only dimension in Africa to exceed the global implementation average.

Most of Africa’s governance frameworks also remain non-binding, Just 21 per cent of the continents’ 170 documented cases of policy areas covered by frameworks throughout the 39 countries are legally enforceable, with the majority taking the form of strategies, guidance or draft policies.

The Index also finds that civil society efforts are largely focused on building awareness, strengthening capacity and fostering collaboration, while fewer initiatives concentrate on accountability and oversight. At the same time, documented cases of unacceptable-risk AI in Kenya, Ghana and Uganda highlight the urgent need to strengthen governance as AI adoption accelerates across the continent.

For more conversations on technology in the public sector, register for Smart Gov KE.