António Guterres’s warning about artificial intelligence isn’t surprising, but it does mark a shift in tone. The UN no longer talks about AI as a promising technology to be accompanied by ethical principles. Instead, it refers to an infrastructure capable of disrupting economies, employment, elections, security, and power dynamics. And it does so with very clear urgency: AI is advancing faster than the regulations that should guide it.
The UN Secretary-General opened the first Global Dialogue on AI Governance with a phrase that succinctly captures the moment: artificial intelligence is progressing at an uncontrollable pace. According to Guterres, it is deploying faster than even those building it can follow, which is why he calls for global rules, safeguards, and international cooperation.
The event, held in Geneva on July 6-7, 2026, aims for a challenging goal: to lay the groundwork for an international conversation on AI that isn’t dominated solely by leading companies or countries with the most computational capacity. The UN itself envisions a second session in New York in May 2027, illustrating that the process has just begun and won’t immediately result in a binding treaty.
The problem is no longer just technical
For years, much of the debate about AI focused on biases, privacy, copyright, employment, and automation. All those topics remain relevant, but the conversation has moved toward something deeper: who controls systems capable of reasoning, generating code, producing content, operating tools, and acting with a certain degree of autonomy.
The preliminary report from the Independent Scientific Panel on AI, composed of experts from various regions, warns that current safeguards cannot keep pace with the growth of these systems’ capabilities. The UN presents this as the first comprehensive, independent scientific assessment of AI’s opportunities, risks, and impacts.
The technical concern is better understood when examining model evolution. In a short period, models have gone from answering questions to writing software, analyzing extensive documents, creating videos, navigating interfaces, connecting to APIs, and becoming part of agents that perform chained tasks. The risk is no longer just a chatbot giving an incorrect answer but systems making decisions, manipulating information, or acting within critical processes.
Reuters summarized the UN panel’s warning across several fronts: misinformation, cyberattacks, biological threats, deceptive behaviors, loss of human control, and agentic systems capable of performing complex tasks autonomously. These risks vary in likelihood and timeframe but share a common feature: they escalate more quickly when models are linked to real-world tools.
Children—the late-discussed limit
Guterres has emphasized the protection of minors. The UN proposes that companies demonstrate their products are safe for children before deployment, establish zero tolerance for child sexual abuse generated or facilitated by AI, and consider age limits or controls similar to those applied to other sensitive products.
This is one of the least controversial points politically. A system capable of generating images, simulating emotional bonds, recommending content, or engaging in prolonged conversations with minors needs clear rules. Merely including warnings in terms of use or leaving responsibility to families isn’t enough.
From a technological standpoint, the key questions involve implementation. How can age be verified without compromising privacy? How can a model’s behavior be audited when it changes with each update? What measures are needed for open models or those deployed outside major platforms? How can illegal material generation be prevented when weights, fine-tuned models, and services are hosted in different jurisdictions?
Child safety is a legitimate priority, but it also tests the entire regulatory architecture. If the solution ends up being just more identity verification, greater concentration in large platforms, and restrictions on small projects, it could reinforce the dominance of existing market players.
Regulation also means trying not to lose control
There’s a less comfortable aspect to this debate. The UN, governments, and regulators aren’t just concerned about AI risks—they’re also trying to understand what role they might have if the future cognitive infrastructure ends up controlled by a handful of companies, a few countries, and an increasingly capable open-source community.
Cognitive infrastructure concentration explains this well. Guterres highlighted that the United States accounts for about 75% of advanced AI supercomputing, compared to 15% in China. This gap leaves the rest of the world in a secondary position to train, deploy, and govern their own models.
Hence, AI governance isn’t solely about security—it’s about sovereignty. A country reliant on external models, clouds, chips, and platforms can use AI but will have less ability to customize it to their language, laws, public services, education system, or industrial needs.
This uncomfortable intuition is that those in power are now worried about AI because they are unsure how their own roles will evolve. Not necessarily out of a desire to slow progress but because AI could reshape power structures before institutions learn how to govern it.
History repeats with other general-purpose technologies. The internet rolled out before many governments understood its impact on information, commerce, privacy, or propaganda. Social networks were seen as mere communication platforms until they became infrastructures of political influence. With AI, the window for reaction might be narrower because its automation capacity is far greater.
The danger of regulation that favors big giants
The UN calls for global rules, but regulating AI won’t be as simple as listing forbidden practices. Systems are diverse, uses are varied, and architectures evolve rapidly. A closed model used in banking isn’t the same as a small local model for internal support, educational tools, programming agents, or military systems.
The main risk is that regulation could create high entry barriers. Audits, evaluations, red teaming, documentation, traceability, age controls, incident management, and regulatory reporting might be necessary for high-impact systems but also incur costs. Large platforms can absorb these costs, while startups, labs, universities, and open projects will face greater challenges.
Poor regulation could lead to increased concentration of AI power. If that happens, the same actors worrying the UN today could consolidate even more control.
The alternative isn’t to leave everything unregulated. It’s naïve to think that a technology capable of spreading misinformation, automating attacks, or mediating sensitive decisions can grow without limits. Regulation should be risk-based: demanding more where impact is significant, protecting minors and citizens, but avoiding turning every AI experiment into an unmanageable bureaucratic race.
AI won’t wait
The UN is right to try to open a global table. No single country can resolve the risks of a border-crossing technology alone. It also recognizes that governance must include countries that currently lack training capacity or infrastructure because otherwise AI becomes another layer of technological dependency.
But an unavoidable reality is that AI won’t wait for multilateral consensus. Open models keep advancing, agents are starting to integrate into real workflows, companies seek greater automation, and geopolitical pressure from the US, China, Europe, and other blocs pushes for rapid action.
Standards can mitigate harms, impose responsibilities, and set boundaries. But they can’t freeze the pace of change. AI is already transforming software, education, content creation, cybersecurity, scientific research, and how companies organize tasks.
Thus, the real debate isn’t about “regulate or not.” It’s about who governs, with what legitimacy, with what technical capacity, and who benefits. The UN aims to prevent AI from going out of control. Governments want to avoid being left behind. Companies want to keep operating. And users are already adopting these tools even before a proper legal framework exists.
Technology advances, but politics seek to regain control. The question is whether it will succeed without excessively slowing the progress.
Frequently Asked Questions
What did António Guterres say about AI?
He warned that AI is advancing faster than regulations and called for global rules to manage risks, protect minors, and prevent an excessive concentration of technological power.
What is the Global Dialogue on AI Governance?
It is a UN initiative to bring together governments, companies, academia, and civil society around international governance of artificial intelligence. The first session was held in Geneva in July 2026.
Why is the UN concerned about AI?
Due to its potential impact on employment, elections, misinformation, cybersecurity, minors, human rights, international security, and inequality between countries with and without computational capacity.
Can regulation favor big tech companies?
Yes, if obligations are too heavy or uniform. Large companies can bear compliance costs that would be difficult for startups, universities, or open-source projects.
