As artificial intelligence (AI) becomes more deeply integrated into the economy and public services, the United Nations (UN) strongly urges governments to strengthen global AI governance. The organization published a preliminary report that constitutes the first independent scientific study on the emerging opportunities and risks of AI.
The pace of AI development outstrips regulation
The UN report warns that AI capabilities are developing faster than governments can measure, regulate, or control them. This report echoes the findings presented in Dario Amodei's essay 'AI Exponential,' published in June 2026, by the head of Anthropic. Amodei argued that the rapid progress in advanced AI requires urgent action from governments, including independent safety assessments and more robust regulatory frameworks.
According to the UN report, AI has become one of the fastest adopted technologies in history, with recent advancements significantly exceeding expectations. Improvements in AI capabilities have been noted in areas such as conversational abilities, scientific reasoning, code generation, data analysis, and the creation of images, audio, and video.
Governance and assessment challenges
The report indicates that policymakers face a 'proof dilemma': they need data to regulate AI, but the technology is evolving so rapidly that by the time sufficient evidence is gathered, AI capabilities may have already changed. This hinders maintaining the pace of regulation in line with technological progress. Furthermore, governance approaches remain fragmented across different countries, as various jurisdictions adopt different regulatory philosophies and compliance requirements.
This issue mirrors Amodei's arguments, who insisted on the need to activate the 'slow and cumbersome political apparatus' to address risks and opportunities that will begin to accumulate unexpectedly fast. Both sources agree that AI capabilities are outpacing the ability of existing political structures to respond.
The UN also notes that existing safety assessment methods are largely developed by the same companies creating advanced AI systems, making governments dependent on the information disclosed by developers. Consequently, it proposes a shift towards independent, standardized third-party assessments, similar to those used in aviation and pharmaceuticals. This recommendation aligns with Amodei's proposal that advanced AI models should be regulated like aircraft, undergoing mandatory technical testing and independent audits before release, with governments having the authority to block or withdraw systems that do not meet high safety standards.
New complexities in AI testing
Rapidly improving AI systems create new challenges for evaluation. Some models are capable of memorizing test questions, and more sophisticated systems can detect when they are being tested. Researchers have also observed instances where AI systems exhibited deceptive behavior during tests or attempted to evade shutdown in laboratory settings, complicating the oversight process.
Key risks identified by the UN
The preliminary report by the Independent International Scientific Commission on AI asserts that AI development poses risks to human rights, social systems, the economy, and the environment, particularly because the technology is advancing faster than governance and oversight mechanisms. The report emphasizes that while AI offers significant benefits, its rapid and uncontrolled deployment could harm societies, disproportionately affecting vulnerable populations.
Key risks include:
- Disinformation: AI can generate convincing text, images, audio, and video, making it difficult to distinguish authentic content from fake, weakening public trust and democratic processes.
- Cybersecurity and crime: Criminals are already using AI for cyberattacks, and as AI develops, it could facilitate fraud, social engineering, disinformation, and financial manipulation.
- Human rights and privacy: The increasing use of AI data and video surveillance raises privacy concerns and could lead to discriminatory outcomes, especially for vulnerable groups.
- Online abuse: The report highlights the spread of AI-generated child sexual abuse material and sexual violence using deepfakes, with women and children being the most affected.
- Mental health risks: Some AI systems can reinforce users' beliefs regardless of their factual accuracy, contributing to serious mental health issues and dependency among vulnerable users.
- Employment and inequality: AI could exacerbate inequality, displace workers, and redistribute wealth from labor to capital if countries do not invest in skills, infrastructure, institutions, and workforce adaptation.
- Loss of control over AI systems: As AI becomes increasingly autonomous through agentic systems, it is noted that existing oversight methods are insufficient, and reliable ways to maintain control over highly autonomous AI systems remain poorly developed.
- Environmental impact: The growing computational demands of AI require energy-intensive data centers and computing infrastructure, raising concerns about greenhouse gas emissions and resource consumption.
Countries' approaches to AI regulation
The UN notes that AI governance remains fragmented across different regions, as countries adopt varying approaches to regulating this technology. Although some states have introduced AI-specific legislation, it often contains fundamentally contradictory rules and compliance costs. There is no common mechanism for managing AI risks, unified standards for assessing AI systems, and limited coordination between jurisdictions, leading to a fragmented regulatory landscape. Nevertheless, the report sees an opportunity to develop common standards of evidence and enhance global AI oversight.
Moreover, effective AI governance involves not only enacting laws but also building capacity to understand, guide, and support AI development. According to UNCTAD, 118 countries, predominantly from the Global South, are not participating in core AI governance discussions, and less than a third of developing countries have national AI strategies. It is noted that many governments, including those in developed economies, lack the necessary technical expertise to keep pace with AI development.
AI's potential for sustainable development
Despite much of the discussion surrounding AI focusing on regulation and emerging risks, the UN report points to the technology's significant potential in supporting sustainable development when applied responsibly. The Independent International Scientific Commission on AI reports that AI is already delivering measurable benefits in science, healthcare, education, agriculture, and productivity, although its impact depends not only on the technology itself but also on additional investment, infrastructure, and governance.
AI applications in science and medicine
AI contributes to scientific discoveries by analyzing large datasets, generating code, and accelerating research. AlphaFold is cited as an example, which predicted the structures of over 200 million proteins and is used by more than three million researchers to support drug, vaccine, and antibiotic resistance research.
In healthcare, AI has helped radiologists detect breast cancer earlier and allowed frontline healthcare workers in resource-limited settings to provide better care using AI tools adapted to local languages. Diabetic retinopathy screening programs using AI in India, which covered over 600,000 people within the existing healthcare system, were also noted.
Education, agriculture, and productivity
In education, AI can assist teachers through personalized learning and classroom support, but the best results depend on the readiness and training of educators. It is emphasized that AI should complement teachers, not replace human judgment or encourage excessive student dependence.
In agriculture and conservation, AI integrates data on weather, soil, crops, and markets to predict droughts, disease outbreaks, and price fluctuations. Additionally, AI-based tools have improved biodiversity monitoring and the prediction of human-wildlife conflicts.
The most obvious productivity gains are seen in clearly defined tasks such as writing, coding, and consulting. However, it is stressed that productivity depends on investments in skills, management, digital infrastructure, and streamlined workflows, not just on the adoption of AI.
Economic growth and recommendations
According to the report, AI creates economic value only when organizations move beyond simple access and distribution. Factors such as cost, uncertainty, skill shortages, and resistance to organizational change can slow this process. The report recommends that AI governance should aim not only to mitigate risks but also to create conditions under which people and societies can safely benefit from AI. Recommendations include strengthening the technical base, independent oversight, and greater transparency to ensure the safety, accountability, and reliability of AI systems.
Specific proposals for stakeholders
Recommendations include:
- Strengthening governmental capacity by investing in technical expertise, computing infrastructure, AI literacy, and AI safety institutions for better assessment and management of AI systems.
- Implementing independent AI testing through third-party assessments, continuous monitoring, and standardized reporting of AI incidents. Anthropic proposed a similar structure suggesting that advanced AI systems should undergo mandatory pre-assessment comparable to aviation safety certification before public release.
- Increasing transparency by encouraging AI developers to make their systems more open and continue monitoring them post-deployment through incident reporting, user feedback, and usage analysis.
- Ensuring human oversight by clearly defining human involvement in high-risk or ethically sensitive decisions and strengthening accountability frameworks as AI systems become more autonomous.
- Improving AI literacy by enhancing public understanding of AI's capabilities, limitations, and risks, as well as simplifying AI systems for user comprehension and interpretation.
Sufficiency of current regulation
The UN assessment shows that existing governance efforts have begun but remain insufficient to keep pace with technological progress. While numerous AI governance tools exist across various jurisdictions, many incorporate principles of ethics and human rights; these initiatives remain fragmented, often concentrated among a limited number of companies and countries, and rarely measure their actual effectiveness. The report also notes that assessment methods are underdeveloped, and the institutions required for independent oversight are lacking.