Artificial intelligence is delivering significant economic and scientific gains, but its capabilities are advancing faster than governments and institutions can effectively regulate, according to a United Nations-backed scientific panel.
The warning was issued by the UN Independent International Scientific Panel on Artificial Intelligence in a report published Wednesday, which argues that existing governance systems remain fragmented and largely confined to corporate environments. The panel says this fragmentation is increasing the urgency for new platforms that can facilitate structured dialogue between AI developers, governments and the scientific community.
Composed of 40 scientists and experts from around the world, the panel has been tasked with producing the first global independent scientific assessment of AI’s opportunities, risks and broader impacts. Its mandate includes identifying evidence gaps and providing a foundation for policymakers as they design regulatory frameworks.
The report is described as preliminary and is intended to feed into a more comprehensive assessment scheduled for release in 2027. It will also be presented at the United Nations’ inaugural Global Dialogue on AI Governance in Geneva in July, where international coordination on AI oversight is expected to be a central focus.
António Guterres said the report is critical for informing global policy decisions, stating that “the world cannot govern what it cannot understand,” and warning that while the potential benefits of AI are significant, the risks are real and the cost of delay is rising.
The panel’s findings highlight rapid acceleration in AI capabilities driven by increased computing power, advanced training data, and new methodological breakthroughs. These developments are already producing measurable benefits across fields such as medical research, including drug discovery, vaccine development, and cancer detection, as well as applications in agriculture, accessibility, science and information technology.
However, the report stresses that the benefits of AI are not guaranteed and depend heavily on how the technology is managed. It warns that governance capacity is not keeping pace with technological development, particularly in advanced “frontier” systems, where technical expertise is unevenly distributed across countries and institutions.
One of the key challenges identified is what the panel calls an “evidence dilemma,” where policymakers require robust data to regulate effectively, but such evidence often emerges only after technologies have already been widely deployed. This lag, the report argues, may leave governments unable to respond in time to fast-evolving risks.
The panel also highlights growing cybersecurity threats, noting that agentic AI systems capable of operating with minimal human oversight could significantly increase vulnerability to attacks. It cites findings indicating that a large proportion of documented attacks on widely deployed coding agents have been successful, underscoring concerns about system security and oversight.
Beyond cybersecurity, the report warns of broader societal risks, including potential harm to mental health and increased misuse of AI systems for fraud, disinformation and the creation of illegal content. It also raises concerns about the difficulty of conducting reliable safety testing, noting that advanced models may learn to evade evaluation or even deceive testers.
The report further argues that AI’s economic benefits may not be evenly distributed. It highlights disparities between developed and developing countries in access to infrastructure, computing power and governance capacity, warning that such gaps could reinforce existing global inequalities.
While AI is widely expected to boost productivity and economic growth, the panel notes that evidence on labour market effects remains mixed. Some data points to job displacement in certain sectors, while other studies suggest limited impact on employment and wages. The report concludes that it remains unclear whether AI will ultimately create sustainable employment or shift wealth toward capital owners at the expense of workers.
The panel calls for stronger governance frameworks that combine binding regulation with voluntary industry standards, arguing that coordinated global action will be necessary to manage risks while preserving the technology’s potential benefits.
