President Donald Trump commanded the world’s attention when he addressed the United Nations General Assembly on Tuesday. Wars, sovereignty, immigration, and America’s place in the world dominated the headlines. But buried within that sweeping address was a declaration that may outlast everything else he said.

“The United States totally rejects any attempt to construct a globalist scheme of control for the artificial intelligence being spoken of so much now,” Trump told the General Assembly. He then announced that Washington would henceforth call the technology “super intelligence” and pledged to “encourage it, not rein it in.”

The next afternoon, across the U.N. complex, the Security Council heard a very different message.

France, which holds the council’s rotating presidency, convened a high-level briefing devoted to artificial intelligence and international security. OpenAI CEO Sam Altman, Anthropic CEO Dario Amodei, Hugging Face CEO Clément Delangue, and Yoshua Bengio, the pioneering researcher who co-chairs the U.N.’s Independent International Scientific Panel on AI, confronted a question that until recently sounded more like science fiction than international security: What happens when the most capable AI systems pursue goals their creators never intended?

The loudest warnings no longer come from critics outside the AI revolution. They come from the people building it.

Amodei told the council that if AI is managed poorly, it “could be a risk to humanity as a whole.” Altman warned that the moment “calls for extreme care” and that the most consequential decisions cannot rest with a few labs in San Francisco. Bengio was blunter. “The dangers are real and imminent,” he said, describing a threat no single nation can contain.

Bengio’s panel supplied the evidence behind those warnings. Its first thematic brief, released September 21, examined the summer incident in which AI agents under evaluation at OpenAI slipped network restrictions, passed information between runs meant to be kept separate, and compromised OpenAI research infrastructure and Hugging Face’s live systems. The panel stopped short of predicting catastrophe. It concluded instead that three conditions researchers have long feared — a misaligned goal, the capability to pursue it and an environment that permits it — had come together “in a real system, not a laboratory.”

I have written in these pages about the cyber incidents behind that finding. What changed this week is where the finding now sits: in a report from a scientific body created by the General Assembly, placed before the council charged with keeping international peace and security.

The briefers did not agree on the remedy. Bengio has called for licensing and mandatory liability insurance for frontier developers. Amodei spoke days after publishing an essay urging the industry to pace frontier development and open its labs to outside evaluators. OpenAI, in a proposal released two days before the session, urged Washington to lead an international effort on common technical standards and incident reporting, warning that failure to lead would let a “fragmented, uneven, and conflict-ridden system take hold.”

Delangue introduced an essential complication. His company was the target of that OpenAI-agent attack, yet he told the council the greater danger is not powerful AI but the gap between those who have it and those who do not — between attackers and defenders, and between a few companies and everyone else. When closed frontier models refused to help his engineers respond because their safeguards could not tell defenders from attackers, Hugging Face turned to an open-source model built by a Chinese developer. That detail should trouble anyone in Washington who assumes safety and security always point in the same direction.

That is one reason simply telling everyone to slow down is not much of a strategy.

There is another reason: China.

Beijing’s state-backed Global Times dismissed Amodei’s pacing proposal as a “Cold War playbook” designed to preserve American dominance. At the council, China’s representative rejected what he called a “petty us-versus-them clique in the tech sector” while promoting Beijing’s alternative vision of global AI governance. Beijing is also building the institutions to match. In July, 29 countries, among them Russia, Cuba, Venezuela, and Pakistan, signed the founding agreement for the China-backed World Artificial Intelligence Cooperation Organization, headquartered in Shanghai.

Washington’s representative, White House science adviser Michael Kratsios, held the administration’s line, telling the council that international dialogue must not drift toward global governance. He is right. Giving an international body authority over American AI would place consequential decisions partly in the hands of governments whose interests and values often diverge sharply from ours.

But sovereignty is not a vacuum. As I argue in “The New AI Cold War,” this contest turns as much on who writes the rules as on who builds the best model. If America declines to shape the rules, Beijing will gladly write them, and it is already recruiting signatories.

That is why the debate emerging from New York should not be reduced to a false choice between an international bureaucracy controlling American AI and every nation racing ahead without meaningful safeguards. Last week I argued that Trump and Xi need an emergency channel and incident notification, not a pause. Reuters reports the two governments have discussed just such a notification system ahead of today’s summit. That is the right template: American-led technical standards, practical guardrails with rivals and no transfer of sovereign authority to a U.N. bureaucracy.

For Christians, there is an even deeper issue.

Ezekiel’s watchman was charged with sounding the alarm when he saw the sword coming. Scripture then places the burden on those who hear: “then if anyone who hears the sound of the trumpet does not take warning, and the sword comes and takes him away, his blood shall be upon his own head” (Ezekiel 33:4, ESV).

This week the trumpet sounded from an unexpected quarter. Some of the watchmen are also building the technology about which they warn, and their commercial interests deserve scrutiny. But a warning does not become false because the messenger has a stake in the outcome. Once it is heard, responsibility passes to those who heard it: presidents, legislators, business leaders, pastors, and parents.

Trump’s General Assembly speech received most of the world’s attention. Yet the following day’s Security Council meeting may eventually prove just as consequential.

The builders did not agree on the solution, and neither do the nations. But they agree there is a problem, and its consequences will not stop at national borders.

America was right to resist surrendering control of its technological future to an international bureaucracy. But rejecting a bad answer does not answer the warning. Frontier AI has become a global security issue whether Washington wants it to be one or not.

America is already in that debate. Washington should lead it, protecting American sovereignty and technological strength while building practical safeguards with any nation serious about keeping these systems under human control.