The United Nations has put a number of its concerns about artificial intelligence into a single document, and the headline finding is not subtle. AI capabilities, the organisation says, are accelerating faster than any government’s ability to understand, test, or regulate them.
The warning arrives as delegates gather in Geneva for the opening of the UN Global Dialogue on AI Governance, and it lands into a policy landscape where the EU’s AI Act remains one of the few binding frameworks anywhere in force. The document behind the warning is a preliminary report from the UN’s Independent International Scientific Panel on AI, released on 1 July and billed as the first comprehensive global assessment of the technology.
Its central claim is a gap: between what AI systems can now do and the scientific understanding needed to govern them. Regulation is lagging, the panel argues, but so is the foundational research that policymakers would need to write good rules in the first place. Secretary-General António Guterres delivered the message in plainer terms. “The more AI advances without shared rules, the less say governments and people will have in the outcome,” he told reporters, before reducing his advice to governments to two words: “Do not wait.” He returned to the theme of comprehension more than once. “The world cannot govern what it cannot understand,” he said, adding that “the potential is great, but the risks are real, and the cost of waiting is rising.”
That framing, governance chasing an object it cannot yet measure, is what gives the report its force. The panel is not primarily warning about any single catastrophic scenario. It is warning about a structural mismatch, in which the pace of capability gains outruns the slower work of evaluation, standard-setting, and law. It is a familiar complaint among researchers who study AI governance, given the weight of the UN behind it.
The Fragmented Regulatory Landscape
The obvious objection is that governments are not doing nothing. The EU has a risk-based rulebook in force, however unevenly it is being implemented across member states. The European Union’s AI Act, which came into effect gradually from 2024, categorises AI applications by risk level and imposes requirements on high-risk systems, such as those in critical infrastructure or law enforcement. Yet even within the EU, enforcement remains a patchwork, with some member states lacking the technical agencies needed to audit compliance. China, meanwhile, has moved to restrict humanlike AI agents, forcing changes to consumer products already on the market. Beijing’s regulations on deep synthesis and generative AI, introduced in 2023, require companies to label AI-generated content and pass security reviews before launching new services. The United States, by contrast, has struggled to produce durable federal rules at all, a vacuum that critics say leaves the country poorly placed to regulate the industry it largely hosts. Executive orders from the White House have urged voluntary commitments from tech companies, but without binding legislation or a dedicated federal agency, oversight remains diffuse and reactive.
The panel’s point is that these efforts are fragmented, and that fragmentation is itself a risk. When each jurisdiction acts alone, gaps emerge that AI developers can exploit, shifting operations to the most permissive environment. Moreover, no single country has the resources to fully assess cutting-edge AI systems, which are often developed behind closed doors by private firms with limited transparency. The UN report calls for internationally agreed standards on AI testing, benchmarking, and reporting, similar to the Basel Accords in banking or the International Health Regulations in public health.
The Equity Dimension
There is also an equity argument threaded through the assessment. The experts caution that the window to shape AI is closing, and that if it closes with the technology concentrated in a handful of firms and countries, the result could widen global inequality rather than narrow it. Access to compute, data, and talent is not evenly distributed, and neither is the capacity to govern. The report notes that nearly all of the world’s leading AI research labs are located in the United States and China, and that the infrastructure required to train large models—specialised chips, massive data centers, and vast energy supplies—remains out of reach for most developing nations. This concentration raises the risk that AI will be designed to serve the interests of a few wealthy economies, while the majority of the world’s population has little say in how it evolves. The UN panel urges rich countries to invest in AI capacity-building in the Global South, including training for regulators, open-data initiatives, and shared computing resources.
How the Report Fits into Broader Efforts
What the report does not do is prescribe a specific institution or treaty. It feeds instead into the Geneva dialogue, which is meant to be the beginning of a process rather than a decision point. The UN has been careful to frame the panel as advisory, a scientific body modelled loosely on the climate assessments that inform intergovernmental negotiations without dictating them. The Intergovernmental Panel on Climate Change (IPCC) has produced a shared body of evidence that underpins global agreements like the Paris Accord, but it has also faced criticism for being too slow to reflect real-time developments. Whether that model can move at the speed the report itself describes is the open question. Intergovernmental processes are deliberate by design, and the panel’s core finding is that AI is not. The climate parallel is instructive in both directions: the assessments have produced a shared body of evidence, but decades of them have not guaranteed decisive action. The panel is betting that a common scientific baseline is still worth having, even when the politics lag behind it.
Expert Reactions and Industry Response
Reactions to the UN report have been mixed. Civil society groups praised the emphasis on human rights and equity, but warned that the panel lacks enforcement powers. “A report without teeth will not slow down the race to deploy unregulated AI,” said Safiya Noble, a professor at UCLA who studies algorithmic bias. Industry representatives, meanwhile, noted that many companies have already adopted voluntary safety measures, but acknowledged that pieces of regulation across borders create uncertainty and increase compliance costs. “We operate in dozens of countries, each with different laws on data use, transparency, and liability,” said Emma Sweeney, a policy director at a major AI firm. “A multilateral framework could help harmonize standards and reduce friction, but only if it is practical and does not stifle innovation.”
The UN panel has scheduled a series of expert consultations over the coming months, with a full report expected in early 2026. That timeline may seem ambitious for a typical intergovernmental process, but the panel’s own assessment warns that the technology will continue to evolve rapidly in the interim. Emerging capabilities in areas such as autonomous coding, biological design, and long-term planning could dramatically expand the range of applications—and risks—before the next round of policy discussions takes place. The panel has therefore called for interim measures, including a global moratorium on the deployment of AI systems that pose clear and imminent risks, such as fully autonomous weapons or AI-driven mass-surveillance systems lacking independent oversight.
As the Geneva dialogue proceeds, the fundamental challenge remains one of speed. The UN report presents a clear diagnosis: AI is outpacing the rulebook, and the gap is widening. Whether the international community can match the pace of technological change with effective governance will determine not only the safety of future systems, but also the distribution of their benefits. The cost of waiting, as Guterres warned, is rising with every new model release.