The rapid development of artificial intelligence could exacerbate global inequality. This warning comes from a preliminary report by the UN's Independent International Scientific Panel on AI.

The document was prepared by 40 independent scientists and experts. They argue that mere access to these tools does not guarantee equal benefits. Countries reliant on foreign models, cloud infrastructure, and data flows may gain access to technology but lose practical control over its standards, safeguards, and adaptation to local conditions.

The authors highlight that the main issue is the concentration of computational infrastructure and the development of advanced models. According to their findings, the United States controls about three-quarters of the computational power of the world's leading AI supercomputers, while China holds around 15%. Together, these two countries account for approximately 90% of the computations used to train the most powerful systems.

The UN warns that the concentration of AI capabilities among a small number of companies and states could enhance authoritarian control and weaken democratic accountability. The authors note that lagging countries need to develop their own computational and data infrastructure, specialists, and local developers.

“In the long run, the gap between rapidly growing functional capabilities and the availability of effective risk management methods could lead to catastrophic outcomes,” the report states.

A separate section addresses linguistic inequality. According to the researchers' findings, generative models perform better with English and other widely spoken languages, while others are either poorly represented or yield significantly lower quality.

The report cites an example of machine translation from Tigrinya: the system confused smallpox with syphilis, gonorrhea with diabetes, and a phrase about intravenous antibiotics with "intravenous insecticides."

“Such translation errors can pose life-threatening risks,” the authors emphasized.

Another factor contributing to inequality is basic internet access. According to the International Telecommunication Union, 2.2 billion people will remain offline by 2025. For these regions, the implementation of AI first hinges on access to the internet, energy, and digital skills.

The preliminary report is not a regulatory document and does not establish mandatory rules. The panel provides a scientific assessment of the opportunities, risks, and consequences of AI that governments can use when developing policies. Among the areas highlighted by the authors are:

  • development of local AI infrastructure and data centers;
  • increasing AI literacy in schools and workplaces;
  • investment in developers and technical expertise;
  • creation of AI safety institutions;
  • strategies to combat disinformation;
  • evaluation of AI systems after deployment on real users, tasks, and conditions.

The report also pointed out risks related to disinformation, fraud, election interference, human rights violations, and pressure on labor markets.

As a reminder, in June, the Bank for International Settlements concluded that the investment boom surrounding artificial intelligence, which supported the global economy in 2025, is itself becoming a source of macro-financial risks.