Artificial intelligence now touches nearly every part of our daily lives. It helps doctors read medical scans. It decides who gets loans. It flags job applicants for interviews. It even influences which children get extra help in school.
Yet a new report shows that governments are racing to write AI rules but failing to build the systems that actually make those rules work.
The second edition of the Global Index on Responsible AI, released in Cape Town, compares 135 countries. Countries are passing laws and signing international pledges at a rapid clip. But the institutions that enforce those laws, the oversight bodies that monitor compliance, and the transparency rules that let citizens see how AI affects them are not keeping up.
This gap between writing rules and enforcing them is not just a bureaucratic problem. It is a human rights problem. When AI makes mistakes in welfare systems, healthcare, policing, or education, people get hurt. And without proper oversight, those mistakes go unnoticed and unchallenged.
Most Countries Score Below 50 Out of 100
The Global Index on Responsible AI measures countries across five dimensions: AI use in public services, ethics and sustainability, inclusion and diversity, labour and skills, and trust and safety.
The average global score sits at just 35 out of 100. That means even the best-performing countries have plenty of room for improvement. Norway leads the world with a score of 75.26, followed by Italy at 72.71 and Ireland at 71.39. France ranks fourth with 70.32.
But these top scores are the exception, not the rule. Most countries fall far behind. And the gap between the Global North and Global South is stark. Countries in the Global North average 55 points. Countries in the Global South average just 27 points.
Having a Strategy Does Not Mean Having Protection
The report found that 88% of countries in the Global South now have responsible AI content in their frameworks, up from the first edition. That sounds like progress. But 78% of these frameworks are non-binding. They are statements of intent, not enforceable rules.
A country can have a beautiful AI strategy document on a government website. But if that document does not create real obligations, it does not protect anyone. Rachel Adams, founder and CEO of the Global Center on AI Governance, put it bluntly.
Responsible AI cannot be secured through principles alone. Governments need enforceable obligations, independent oversight, public disclosure, monitoring systems and accessible routes for redress.
Evidence of actual implementation exists in only 55% of countries with responsible AI frameworks. In the Global South, that number drops to 45%.
Governments Hide Their Own Use of AI
One of the most troubling findings involves transparency. Governments are happy to talk about regulating private companies. But they are far less eager to disclose how they use AI themselves.
Only 18% of countries require their government to disclose the algorithmic systems they use. That means in most countries, citizens have no way of knowing whether a government algorithm is deciding their welfare benefits, their child’s school placement, or their eligibility for housing.
This lack of transparency creates a blind spot. Governments can hold private companies accountable while operating their own AI systems in the shadows. That double standard undermines public trust and leaves citizens with no recourse when government AI makes mistakes.
Children Are Being Prepared for AI But Not Protected From It
Schools around the world are teaching AI literacy. Children are learning how to use AI tools and how they work. The report found that AI literacy is widespread.
But here is the problem. Only 41% of countries have frameworks that address children’s rights in AI. We are preparing young people for an AI-driven economy, but we are not protecting them from AI-related harms. Those harms include algorithmic bias in educational tools, privacy violations from AI-powered learning platforms, and exposure to harmful content generated by AI systems.
Nigeria stands out as a positive exception. The country ranked 38th globally with a score of 45.93 and was recognised as a global “Bright Spot” for its approach to preparing citizens for AI while strengthening protections. The Nigeria Data Protection Act requires parental consent before children’s personal data can be processed and restricts automated decisions involving individuals.
The Environmental Cost of AI Gets Ignored
AI consumes enormous amounts of energy. Training a single large AI model can emit as much carbon as five cars over their entire lifetimes. Yet only 27% of countries address AI’s environmental footprint in their frameworks. And most of those frameworks are non-binding.
As AI adoption accelerates, so does its energy consumption. Countries that ignore the environmental impact of AI are not just failing to protect the planet. They are also missing an opportunity to build sustainable AI industries that can thrive in a low-carbon future.
Binding Rules Make the Difference
The index shows that countries with enforceable rules perform better across every dimension. Norway, Italy, and Ireland do not just have nice AI strategies. They have independent oversight bodies, mandatory disclosure requirements, and accessible routes for citizens to challenge AI decisions.
Countries that rely only on voluntary frameworks and non-binding principles fall behind. They have the same ambitions but none of the tools to achieve them.
This pattern holds across every region. In Africa, only six out of 39 countries scored above the global average: Nigeria, Egypt, Kenya, Ghana, Benin, and Morocco. These countries share one thing in common. They have moved beyond principles to create actual enforcement mechanisms.
The Cost of Inaction Is Rising
AI is not waiting for governments to catch up. Generative AI tools are now used by 53% of the global population. AI systems are embedded in welfare systems, healthcare, policing, education, finance, and workplaces.
Every day that passes without proper oversight is a day when AI systems can make mistakes that harm real people. The report found misuse of AI by governments in 35 countries. Those are not hypothetical risks. They are happening right now.
The good news is that countries do not need to start from scratch. They can learn from the top performers. They can build independent oversight bodies. They can mandate transparency for government AI use. They can create accessible routes for citizens to challenge AI decisions. They can turn their non-binding principles into enforceable rules.
A Foundation for Growth and Stability
This push for enforcement is ultimately good for the tech industry. Clear, enforceable rules give companies certainty. When companies know what is expected of them, they can invest with confidence. They can build products that comply with regulations rather than guessing what might be required later.
Enforceable rules also build public trust. When people know that AI systems are being watched and that they have recourse if something goes wrong, they are more willing to embrace AI. That trust is essential for the long-term growth of the AI industry.
The alternative is a fragmented mess. Different countries with different voluntary standards. Companies struggling to comply with conflicting expectations. Citizens are losing faith in both AI and the governments that fail to regulate it. That scenario benefits no one.
The Window Is Closing
The second edition of the Global Index on Responsible AI sounds an alarm. Governments are moving in the right direction by adopting AI strategies and signing international commitments. But they are moving too slowly on the hard part: building the institutions, enforcement tools, and transparency mechanisms that actually protect people.
The report concludes that responsible AI governance is expanding worldwide but remains fragmented, under-enforced, and insufficiently grounded in public accountability.
That is not a sustainable path. As AI becomes more powerful and more embedded in every aspect of life, the cost of inaction will only grow. Governments that act now to build enforcement systems will protect their citizens, build public trust, and create the stable environment that the AI industry needs to thrive.
The rules are being written. The question is whether anyone will enforce them.
The full report is available for free download. You can access the 2026 edition at www.global-index.ai.
You can also visit the Global Center on AI Governance at www.GlobalCenter.AI to learn more about their work and access additional resources.
The data is licensed under Creative Commons, so you are free to share and remix the content for non-commercial purposes as long as you attribute the Global Center on AI Governance and the Global Index on Responsible AI
