Interestingly, the report echoes findings highlighted by Anthropic chief executive Dario Amodei in his essay “AI Exponential”, published in June, 2026. In the essay, he argued that rapid advances in frontier AI require urgent action from governments, including independent safety evaluations and stronger regulatory frameworks.
While the UN report takes a broader scientific and policy-oriented approach, it echoes several of the same concerns raised by Anthropic.
Whether it is Anthropic’s essay or the UN’s assessment, there is a growing chorus of calls for responsible AI development.
AI is moving faster than regulation
According to the UN report, AI has become one of the fastest-adopted technologies in history, with recent advances significantly exceeding expectations for technology development. It notes that AI capabilities have improved across areas including conversational ability, scientific reasoning, code generation, data analysis, and image, audio and video generation.
The report says policymakers face an “evidence dilemma” because they need evidence to regulate AI, but the technology evolves so quickly that by the time sufficient evidence is available, AI capabilities may already have advanced. This makes it difficult for governance and regulation to keep pace with technological progress. It adds that governance approaches across countries remain fragmented, with different jurisdictions adopting different regulatory philosophies and compliance requirements.
This concern closely mirrors Amodei’s recent AI Exponential essay, in which he argued that “we now, globally and collectively, need to activate a slow and rickety policy apparatus to deal with risks and opportunities that are going to compound surprisingly quickly from here.”
Both argue that AI capabilities are advancing faster than existing policy frameworks can respond.
The UN report also notes that current safety evaluation methods are largely designed by the same companies developing frontier AI systems, leaving governments dependent on information that developers choose to disclose. It argues that AI should move towards independent, standardised third-party assessments, drawing parallels with industries such as aviation and pharmaceuticals, where external safety verification is mandatory.
The recommendation aligns with Amodei’s proposal that frontier AI models should be regulated much like aircraft, undergoing mandatory technical testing and independent audits before deployment, with governments empowered to block or withdraw systems that fail to meet high safety standards.
According to the report, rapidly improving AI systems also create new challenges for evaluation. Some models can memorise benchmark questions, while increasingly capable systems may recognise when they are being tested. The report further notes that researchers have observed instances of AI systems exhibiting deceptive behaviour during testing or attempting to avoid shutdown in laboratory settings, making oversight more complex.
What risks does the UN identify
The Preliminary Report of the Independent International Scientific Panel on AI says AI development is creating risks across human rights, social systems, economies and the environment, particularly as the technology advances faster than governance and oversight mechanisms. The report stresses that while AI offers significant benefits, its rapid and unchecked deployment could expose societies to harms that disproportionately affect vulnerable populations.
The report identifies the following key risks:
-
Misinformation: AI can generate convincing text, images, audio and video, making it harder to distinguish genuine content from fake. The report says this could weaken public trust and democratic processes. -
Cybersecurity and crime: The report says criminals are already using AI for cyberattacks. As AI advances, it could also enable fraud, social engineering, disinformation and financial manipulation. -
Human rights and privacy: According to the report, AI’s growing use of data and surveillance raises privacy concerns and could lead to discriminatory outcomes, particularly for vulnerable groups. -
Online abuse: The report highlights the spread of AI-generated child sexual abuse material and deepfake-enabled sexual violence, with women and children among those most affected. -
Mental health risks: The report says some AI systems may reinforce users’ beliefs regardless of their accuracy, contributing to serious mental health incidents and encouraging dependency among vulnerable users. -
Employment and inequality: The report says AI could widen inequality, displace workers and shift wealth from labour to capital if countries fail to invest in skills, infrastructure, institutions and workforce adaptation. -
Loss of control over AI systems: As AI becomes increasingly autonomous through agentic systems, the report warns that existing oversight methods are inadequate. It says reliable methods to retain control over highly autonomous AI systems remain underdeveloped. -
Environmental impact: The report notes that AI’s growing computational demands require energy-intensive data centres and computing infrastructure. It identifies greenhouse gas emissions and resource use as key environmental concerns.
How are countries approaching AI regulation?
The UN says AI governance remains fragmented across regions as countries adopt different approaches to regulating the technology. According to the report, some countries have introduced AI-specific legislation, but with fundamentally contradictory rules and compliance costs. It adds that there is no common mechanism for managing AI risks, no shared standards for evaluating AI systems, and limited coordination across jurisdictions, resulting in a fragmented regulatory landscape. However, the report says there is still an opportunity to develop shared evidence standards and strengthen global oversight of AI.
What opportunities does AI offer alongside the risks?
While much of the debate around AI centres on regulation and emerging risks, the UN report says the technology also has significant potential to support sustainable development when deployed responsibly. According to the Independent International Scientific Panel on AI, AI is already delivering measurable benefits across science, healthcare, education, agriculture and productivity, although its impact depends on complementary investments, infrastructure and governance rather than technology alone.
The report highlights the following opportunities:
Advancing scientific research: AI is supporting scientific discovery by analysing large datasets, generating code and accelerating research. The report cites AlphaFold, which has predicted the structures of more than 200 million proteins and is being used by over three million researchers to support drug discovery, vaccine development and antibiotic resistance research.
Improving healthcare: The report says AI has helped radiologists detect breast cancer earlier and enabled frontline health workers in low-resource settings to provide better care using AI tools adapted to local languages. It also highlights AI-assisted diabetic retinopathy screening programmes in India that examined more than 600,000 people as part of an existing healthcare network.
Supporting education: The report says AI can assist teachers through personalised learning and classroom support, but better outcomes depend on teacher preparedness and training. It notes that AI should complement teachers rather than replace human judgment or encourage excessive reliance by students.
Helping agriculture and conservation: AI can combine weather, soil, crop and market data to forecast droughts, disease outbreaks and price shocks. The report also says AI-driven tools have improved biodiversity monitoring and human-wildlife conflict prediction.
Boosting productivity: The report finds the clearest productivity gains in well-defined tasks such as writing, coding and consulting. However, it stresses that productivity depends on investments in skills, governance, digital infrastructure and redesigned workflows, not AI adoption alone.
Driving economic growth through adoption: According to the report, AI creates economic value only when organisations move beyond access to adoption and diffusion. Factors such as costs, uncertainty, lack of skills and resistance to organisational change can slow this process.
What does the UN recommend for governments, businesses and developers?
The report says AI governance should focus not only on reducing risks but also on creating the conditions for people and societies to benefit from AI safely. It recommends stronger technical capacity, independent oversight and greater transparency to ensure AI systems remain safe, accountable and trustworthy.
Recommendations include:
-
Strengthen government capacity by investing in technical expertise, computing infrastructure, AI literacy and AI safety institutes so governments can better assess and govern AI systems.
-
Introduce independent AI testing through third-party evaluations, continuous monitoring and standardised reporting of AI-related incidents. Anthropic has proposed a similar framework, suggesting frontier AI systems should be subject to mandatory pre-deployment evaluations, comparable to safety certification in aviation, before they are released publicly.
-
Improve transparency by encouraging AI developers to make systems more transparent and continue monitoring them after deployment through incident reporting, user feedback and usage analysis.
-
Ensure human oversight by clearly defining human involvement in high-risk or ethically sensitive decisions and strengthening liability frameworks as AI systems become more autonomous.
-
Promote AI literacy by improving public understanding of AI’s capabilities, limitations and risks while making AI systems easier for users to understand and interpret.
Are today’s AI regulations sufficient?
The UN’s assessment suggests that existing governance efforts have begun but remain insufficient to keep pace with technological progress.
According to the report, dozens of AI governance instruments already exist across jurisdictions, many incorporating ethics and human rights principles. However, these initiatives remain fragmented, are often concentrated among a limited number of companies and countries, and rarely measure their real-world effectiveness. The report also says evaluation methods remain underdeveloped, while the institutions needed for independent capability and risk assessments are still in their early stages.
The report concludes that governments must strengthen technical expertise, improve AI evaluation and build governance systems that evolve alongside AI rather than reacting after new capabilities emerge. That conclusion broadly echoes Anthropic’s warning that policymaking can no longer remain reactive. While the UN frames the issue as a global governance challenge and Anthropic outlines specific regulatory proposals, both agree that the pace of AI development is now outstripping the world’s ability to oversee it safely.
