A new United Nations report warns that the rapid development and uneven adoption of artificial intelligence risk exacerbating global inequality, as a handful of nations and firms tighten their grip on the technology’s most powerful models and infrastructure.
“The more AI advances without shared rules, the less say governments and people will have in the outcome,” said António Guterres, the UN secretary general, at a press conference on Wednesday. “Our message to governments is simple: do not wait … the science is here. We can no longer say we did not know what we do.”
The preliminary report, released by the Independent International Scientific Panel on AI — a body of 40 experts established by the UN General Assembly last year — warns that current safeguards cannot keep pace with the growth of AI capabilities. Policymakers need scientific evidence to govern effectively, the panel argues, but by the time the evidence is clear it may be too late to act.
Deepening digital divide
While more than a billion people now use AI every week, access and usage vary dramatically. Data from Microsoft covering the second half of 2025 shows that 24.7% of the working-age population in the Global North uses AI tools, compared with just 14.1% in the Global South. Adoption in the Global North grew nearly twice as fast, widening the gap rather than closing it.
The concentration of development and investment is even starker. The United States and China dominate the creation of leading AI models and the compute infrastructure — the hardware, memory, networking and storage needed to run powerful systems. Estimates suggest the US controls roughly three-quarters of the computing power of the world’s leading AI supercomputers, with China holding about 15%. Together, the two countries account for around 90% of the computations used to train the most powerful models. Only 32 countries host AI-specialised data centres, overwhelmingly in the Global North; Africa and Latin America together account for just 3% of global AI compute capacity.
“Access to AI tools alone does not produce equal benefit,” the report states. “Countries that rely on foreign models, cloud infrastructure and data pipelines may gain access to AI while losing practical control over its standards, safeguards and local fit.”
The panel warns that this concentration can enable authoritarian capture. “The concentration of AI capabilities in a small number of firms and countries could enable authoritarian capture and undermine democratic accountability,” the report says. AI-enabled authoritarianism can scale globally and operate automatically, influencing information, behaviour and economic coercion.
Most developing countries lack the infrastructure and expertise to benefit. The report notes that “most countries, including many advanced economies, lack the technical expertise to assess the most capable ‘frontier’ models or to participate meaningfully in their governance.”
Linguistic and environmental costs
Inequality also manifests in language. Over 7,000 languages are spoken worldwide, yet most AI models are trained on only around 100. English dominates because of the abundance of online data. “Artificial intelligence leaves most languages behind,” the report states. Generative AI tools perform well in English and other widely used languages, but most languages are either excluded or perform much worse.
These disparities can have life-threatening consequences, particularly in healthcare. The report cites a machine translation from Tigrinya that confused smallpox with syphilis, gonorrhoea with diabetes, and rendered “you have been given intravenous antibiotics” as “you have been given intravenous insecticides.” “These mistranslations can be life-threatening,” the report notes.
Beyond linguistic exclusion, more than 2 billion people — almost a third of the world’s population — remain completely offline, according to the International Telecommunication Union.
The environmental toll of AI is also unevenly distributed. Data centres account for a substantial portion of electricity consumption; AI-optimised data centres are projected to more than quadruple their demand by 2030, often drawing on fossil fuels. Training a single AI model can emit as much carbon as five cars over their lifetimes. Water-based cooling systems consume vast amounts of water — roughly half a gallon per kilowatt-hour. The rapid replacement of servers and GPUs generates millions of tonnes of e-waste annually, with generative AI alone contributing an estimated 1.2–5 million additional tonnes per year.
Risks and opportunities
The report acknowledges AI’s transformative potential in agriculture, education, science and health, but stresses that access alone does not guarantee benefit. “Access to AI tools alone does not produce equal benefit,” it reiterates. Workers fluent in English may advance, while others face technological barriers, widening economic inequality. The lack of linguistic diversity can introduce biases reflecting Anglo-centric and North American viewpoints.
AI also poses risks to democracy through fraud, election interference and the spread of misinformation. The report warns that AI systems can shape public opinion through recommendation algorithms and content moderation, and can be used for behavioural manipulation and economic coercion.
Maria Ressa, co-chair of the panel and a journalist, stressed at the press conference that AI’s “pace is not slowing, the power is concentrating, and control is not guaranteed.”
Proposed solutions
The report functions as a toolkit for UN member states, offering broad guidance rather than binding policy. Suggestions include developing local AI infrastructure such as data centres, which requires securing reliable energy supplies; improving AI literacy in schools and the workforce; investing in developers; building AI safety institutes; creating strategies to combat disinformation; and continuously measuring how AI systems behave after release — “with real users, real tasks and real environments.”
The panel advises countries lagging behind to consider significant investment in computing and data infrastructure, while acknowledging the environmental costs. It also calls for directing resources towards multilingual research and development to address the AI language gap.
The report is intended to inform the inaugural Global Dialogue on AI Governance, which the UN will host on 6 and 7 July 2026 in Geneva, with a second session in New York in May 2027. The dialogue aims to ensure governance reflects the priorities of all nations, not just the most technologically advanced. A 2024 UN report found that 118 countries were excluded from existing AI governance initiatives, with only seven developed nations participating in all frameworks.
When pressed by journalists on why the panel did not recommend an international body to test powerful AI models before public release, the panel pushed back. “It may sound like a cop out that we’re not making policy recommendations, but frankly, when you have the scientists together in the room, they tell you what they know,” Ressa said. “That’s why it is usable in Washington and Beijing and Manila. The prescribing happens next week in Geneva … where the states sit at the table.”
