U.S. blocks global superintelligence governance at the UN, while OpenAI and Anthropic call for cooperation

By Zendoric · September 24, 2026.

The United States rejected before the UN Security Council any attempt to create a global framework to control superintelligence, the term Donald Trump used in his speech to the UN to refer to artificial intelligence. The rejection came from Michael Kratsios, director of the White House Office of Science and Technology Policy (OSTP), as reported by Anadolu Agency.

Kratsios was explicit: international dialogue on AI at the Security Council and in other forums must not lead to global governance. In his view, the United States —which he described as the leader in advanced superintelligence— is already developing the technology safely and responsibly, and there is no reason to pause its progress or to tie it to new international control structures.

Instead, he called on each country to legislate on its own: national laws and rules rather than a global system. The underlying argument, according to his remarks, is that AI's potential will be realized through «open cooperation», sovereign infrastructure and each nation's development of its own capabilities, not through a supranational body of rules.

The detail that weighs most in this piece is not only Washington's position, but the contrast within the same session. Earlier, at that same meeting and according to the same report, the top executives of OpenAI and Anthropic had called for precisely the opposite: that no single actor dominate the development of artificial intelligence and that there be international cooperation to manage its risks. Two statements, the same room, opposing theses on who should set limits and how.

In general, the UN has spent years trying to weave some kind of common floor against the risks of dual-use AI, with modest results: declarations of intent, safety summits, advisory bodies with no binding power. Kratsios's response is not an abrupt novelty either: it fits the line this Administration has been setting in its foreign technology policy, with export controls on advanced semiconductors and a push for «sovereign AI», prioritizing competitive advantage over ceding regulatory ground. What this intervention adds is the most direct formulation so far: it is not just «a treaty is not needed», it is «that framework must not exist».

Our reading is that this divergence between government and labs matters more than it seems at first glance. When the technology's own creators —who have every incentive to avoid demanding regulators— explicitly ask that no one dominate the sector, and their own government responds by shielding precisely the possibility that such dominance be exercised with no counterweight other than whatever law each country chooses to give itself, the gap is not cosmetic. It is a clash between the U.S. interest as a power (keeping its edge and its room to maneuver) and the declared interest of its own companies (spreading the reputational and operational risk of governing something this powerful alone).

It also fits the geopolitical board we have been following in these pages: without a multilateral framework, the race with China is decided country by country, and «sovereign regulation» works in practice as ammunition for not ceding ground to Beijing, which likewise has no incentive to accept external limits. The short-term result is more race and less shared safety net: exactly the scenario in which the odds rise of errors or misuse that no one is coordinating on how to prevent together. This is not catastrophism, it is the arithmetic of incentives: the fewer actors that accept common rules, the more chance weighs on whether the first serious incident involving a very advanced system happens under a regime capable of responding or under one that could not even agree on the vocabulary.

In the long run we continue to think that AI's trajectory toward abundance and better health does not depend on there being a treaty signed in New York, but on capability continuing to advance without an accident that halts it abruptly. That is why this news does not settle the «governance yes or no» debate: it gives a clue as to how the next decade will be played out, with sovereign actors competing, labs asking for guardrails their own governments are not willing to install, and real safety depending, once again, on what each company decides to do behind closed doors.