AI CEOs Begged the UN Security Council. The US and China Said No.

The Day AI CEOs Knocked on the Security Council Door

On September 23 at the UN headquarters in New York, two people who normally operate far from traditional geopolitics — Sam Altman of OpenAI and Dario Amodei of Anthropic — stood before the UN Security Council and told its members that artificial intelligence could threaten all of humanity.

Then the council’s two most powerful members effectively told them to go home.

The meeting was convened by France, which holds the Security Council presidency for the month. Foreign ministers from Britain and France echoed the warnings, pointing to the specific danger of autonomous lethal weapons systems that could select and engage targets without meaningful human control. The framing was sober, the stakes were painted in existential terms, and for a few hours it looked as though the post-WWII body charged with maintaining international peace might actually move on AI.

It did not.

Michael Kratsios, the science and technology adviser to President Trump, delivered the American position with blunt clarity: the United States would “resolutely reject” any attempt to build a framework controlling AI. He also signaled opposition to slowing the pace of development, even for safety reasons. China’s representative offered a slightly more nuanced rebuttal — acknowledging some understanding of regulation while pushing back against Western dominance of the conversation and insisting the UN itself, not a handful of European capitals, should be the primary forum. Russia stayed on the sidelines, arguing the issue belonged in existing specialized agencies rather than the Security Council.

The optics were striking. The CEOs of the two most influential AI companies in the world had just asked the most powerful geopolitical body on Earth to help them prevent a catastrophe. The body’s permanent members could not even agree on whether to discuss it.

Why This Meeting Matters More Than It Looks

The UN Security Council has, since 1945, dealt with threats to international peace — wars, sanctions, peacekeeping missions, nuclear proliferation. It has never before been asked to consider AI regulation. That this question landed on its docket at all signals how seriously European powers now view the technology’s strategic implications.

France’s push was not accidental. European governments have spent the past two years trying to position the EU AI Act and related diplomatic initiatives as the global standard. Bringing the question to the Security Council was a move to elevate AI governance from a regulatory regime to a security mandate — one that could, in theory, carry the force of binding resolutions under Chapter VII of the UN Charter.

That is precisely what alarmed Washington and Beijing.

The United States has long preferred voluntary industry frameworks and market competition to top-down regulation. Kratsios’s language — “resolutely reject” — was unusually sharp for a multilateral setting, suggesting the Trump administration sees AI regulation not merely as a domestic policy dispute but as a strategic question. A Security Council framework, even a non-binding one, could create precedents that constrain American AI companies or legitimize extraterritorial oversight.

China’s position is different but points in the same direction. Beijing opposes Western-led rulemaking not because it favors deregulation — China has its own comprehensive AI governance apparatus — but because it resents that the agenda is being set in London and Paris rather than in New York, where China wields more influence as a permanent Security Council member.

The convergence between the US and China on this issue is rare and significant. They disagree on everything from tariffs to Taiwan to human rights. But both see Western-dominated AI governance as a threat: the US because it fears regulation will cede technological leadership, China because it fears regulation will codify Western advantage under the guise of neutrality.

Who Wins, Who Loses

The winners from September 23’s deadlocked session are the AI companies themselves.

No binding resolution. No mandatory safety standards imposed from above. No requirement to slow development. OpenAI, Anthropic, and every other lab operating at the frontier can continue building at the pace they choose, answerable to no international body and pressured only by the somewhat vague force of their own public statements.

That is exactly how they wanted it.

The losers are anyone who believes the current trajectory carries unacceptable risk. Amodei was explicit: “no single leader, company, or nation can address this alone.” The logic is sound. An AI system developed in Silicon Valley can be fine-tuned in Shanghai, deployed in Lagos, and cause cascading failures across financial systems spanning three continents. The risk is inherently borderless. The governance architecture remains stubbornly territorial.

The European foreign ministers who championed the Security Council session now face an awkward choice. They can retreat to the EU AI Act and hope other countries voluntarily align — a strategy that has worked modestly for data privacy but may prove insufficient for a technology whose most dangerous applications cannot be contained within any single jurisdiction. Or they can try again, perhaps through the broader UN General Assembly, where the power dynamics are less paralyzed but the enforcement mechanisms are thinner still.

What Happens Next

The most likely near-term path is fragmentation. Expect the EU to deepen its regulatory perimeter and attempt to export its standards through trade agreements and procurement rules — the so-called Brussels Effect. Expect the US to double down on innovation-first policy, with government funding flowing toward frontier labs under the assumption that leadership is the best safety measure. Expect China to continue building its own closed ecosystem of AI governance, emphasizing state control and social stability over the openness European and American regulators typically prioritize.

None of these tracks create the kind of global coordination Amodei and Altman requested. But they also do not require the kind of agreement the Security Council proved incapable of delivering.

The deeper question is whether the AI companies will eventually ask for regulation themselves — or whether they will continue lobbying for access to decision-makers while opposing any framework that constrains them. Altman’s September 23 testimony was notable for what it implied: that the companies building the most powerful systems in the world recognize they cannot manage the risks alone. But recognition is not the same as surrendering control.

France will try again. The Security Council presidency rotates monthly, and the next holders — likely a non-permanent member — may be more amenable to the European agenda. But without US participation, any resolution will lack legitimacy. Without Chinese buy-in, it will lack universality.

The UN Security Council convened to discuss the most consequential technology of the 21st century. It produced nothing. That outcome, paradoxically, is itself a result — one that reveals where the real centers of AI governance actually lie, and how difficult it will be to change that arrangement.