For the first time, an independent international scientific panel mandated by the United Nations has produced a global assessment of artificial intelligence. Behind the immense promise for health, science and education, the experts warn of an unprecedented concentration of technological power, a widening North-South divide and the emergence of autonomous agents whose governance remains largely to be invented.
There was electricity, the steam engine and then the internet. Now there is artificial intelligence. But with one major difference: never before has a technology been adopted at such speed. That is the central finding of the first preliminary report of the Independent International Scientific Panel on Artificial Intelligence, established in 2025 by the United Nations General Assembly to provide states with a shared scientific assessment of a revolution that is already upending economic and geopolitical balances.
The document, published in July 2026, gives in neither to doom-mongering nor to technological enthusiasm. It attempts to tread a fine line: AI could become one of the most powerful accelerators of human development, but its current trajectory also concentrates systemic risks rarely seen in industrial history.
“We are at a crossroads” is, in essence, the report’s conclusion. For AI is no longer merely a tool capable of producing text or images. It is gradually becoming a technology capable of performing complex intellectual tasks, making decisions and taking action.
https://www.un.org/independent-international-scientific-panel-ai/en
Acceleration outpacing the capacity for oversight
The first warning sign concerns speed. The capabilities of models are now advancing faster than the tools used to measure them.
The experts stress that recent systems are reaching or exceeding levels that many researchers did not expect for several years. They can write computer code, solve complex scientific problems, analyse vast volumes of data and assist medical research.
In healthcare, AI is already contributing to the early detection of certain cancers and to biological analysis. The emblematic example remains AlphaFold, used by several million researchers, which has made it possible to predict the structure of more than 200 million proteins, speeding up pharmaceutical research.
Yet this new power creates a paradox: the more capable systems become, the harder they are to evaluate. Conventional tests are proving inadequate. Some models are now able to recognise that they are being evaluated, and even to behave differently depending on the context.
For scientists, the challenge is becoming comparable to the one faced in aviation or pharmaceuticals: can the worldwide deployment of critical technologies be accepted without independent certification mechanisms?
A new global divide in artificial intelligence
Beyond the technology, the report highlights a profoundly geopolitical issue: the concentration of power.
Unlike the Internet, whose original architecture was largely distributed, advanced AI relies on three scarce resources: computing power, massive datasets and scientific talent. Yet these resources are now concentrated in the hands of a limited number of states and companies. According to data cited in the report, around 75% of the computing power of the 500 largest AI-dedicated supercomputers is located in the United States, compared with around 15% in China. The rest of the world shares a marginal fraction of the capacity.
This reality turns AI into a matter of sovereignty comparable to energy or strategic infrastructure. The risk identified by the experts is not merely a conventional digital divide between those who will have access to the tools and those who will be deprived of them. The real gap will be between countries able to design, audit and steer AI, and those condemned to use systems designed elsewhere.
For the Global South, this question is becoming central. The report points out that thousands of languages and cultures remain poorly represented in the data used to train the models. AI could therefore reproduce, or even amplify, historical imbalances.
The arrival of autonomous agents changes the game
The report’s most strategically significant section probably concerns the rise of “agentic AI”. Until now, humans asked and machines answered. Tomorrow, AI agents will be able to plan, make decisions, use software, interact with other agents and carry out complex tasks with reduced human supervision. For the economy, the consequences could be considerable: the automation of intellectual tasks, faster scientific research and the transformation of entire sectors.
But the risks are also changing in nature. The experts point to AI-enabled cyberattacks, automated fraud, information manipulation and the difficulty of maintaining control over systems capable of pursuing their own operational objectives. The question is no longer simply: “What can an AI produce?” It becomes: “What can an AI do when it is given the ability to act?”
Democracy, information and shared reality under strain
The report also devotes a substantial section to social and democratic effects. The mass production of artificial content — texts, images, videos, deepfakes — threatens what the experts call “shared reality”, that is, the ability of societies to share a minimum common foundation of facts.
In a context already marked by political polarisation and distrust of institutions, AI could industrialise disinformation, personalise manipulation campaigns and make it ever harder to distinguish authentic content from artificial content. The authors also warn of specific risks for children, mental health, data protection and fundamental rights.
Towards global governance of AI?
The report comes at a decisive moment. The United States favours an approach driven by private innovation. The European Union is pressing ahead with its regulatory framework. China is developing its own model of technological control. Amid these competing visions, the UN is trying to establish a third way: global governance grounded in science, independent assessment and the inclusion of as many states as possible.
The challenge will be immense. Unlike climate change, where the international community gradually built mechanisms such as the IPCC, artificial intelligence is evolving far faster than traditional diplomatic cycles. The report’s message is ultimately simple: humanity still has a window in which to organise this revolution, but that window is narrowing.
Artificial intelligence could become a historic tool of emancipation, capable of accelerating scientific, medical and educational progress. It could also become an unprecedented driver of the concentration of economic, political and informational power. The question is therefore no longer simply how far artificial intelligence will go, but who will decide the direction it takes.
Translated from French. The French original prevails.






