Sam Altman addresses UN Security Council on AI

OpenAI CEO Sam Altman gave an AI speech on September 23 in the chamber of the United Nations Security Council in New York. His central request was a mechanism to build national and international frontier AI standards together for measuring capabilities, assessing risks, determining whether safeguards are sufficient and preserving meaningful human oversight. "We have unilaterally slowed down in the past. We will do so in the future," he said, adding that beating companies in a competitive race is not a reason to make rash decisions. OpenAI published the full text of the remarks the same day.

The meeting was convened by France, which holds the Council presidency for September, and chaired by French foreign minister Jean-Noël Barrot. According to the UN, the briefers were Yoshua Bengio, co-chair of the UN Independent International Scientific Panel on AI, Altman, Anthropic CEO Dario Amodei and Hugging Face CEO Clément Delangue. According to reports, Altman was the second briefer, and the meeting was the Council's first to focus specifically on the safety risks of increasingly capable AI. No written agreement was expected to come out of it.

Altman began by laying out a choice. "AI can either be more like a new Renaissance of creativity and discovery, or more like a new Industrial Revolution of upheaval and disarray," he said. He then noted that some of the things people building AI have said are so dystopian they sound like the plot of bad science fiction movies. He said he largely agreed with the assessment of Bengio, who spoke before him, but saw two dangers rather than three: losing control of the future to AI, and power concentrating in too few hands.

On the first danger he dismissed the debate over probabilities. Whether people put the risk of catastrophe at 10%, 1% or 0.1%, no level is acceptable, and models should not be trained unless there is an extremely strong case that they can be kept under human control. He also said extreme care is needed because AI progress could accelerate rapidly as systems approach recursive self-improvement, the stage at which they improve themselves. On the second danger he said "no one person or company or country should be able to use the most powerful AI models to impose their worldview on everyone else."

OpenAI set out three principles. People must stay at the center of AI decision-making, the benefits of scientific progress and economic growth must be by people and for people, and the technology must empower individuals. "We might be one of the builders of this technology, but we are not the heroes of this story," Altman said. He also cited progress in math: models that were okay at grade-school math three summers ago reached gold-medal level at the most prestigious international math competition last summer, and a few weeks ago one solved the Navier-Stokes equations, a Millennium Prize Problem, according to the claim. According to reports, that result has not yet been independently verified.

The proposal has four parts: complementary national and international frontier AI standards, common standards so countries can compare evidence and verify compliance, accurate and speedy incident reporting and classification protocols so the world can learn from failures before they become catastrophes, and secure channels among governments, critical infrastructure operators and technical experts to share emerging vulnerabilities and threats. He said the standards should not lock in incumbents or favor one business model, and must support open and closed model developers as well as new entrants and established labs. METAL has reported on the frontier AI international standards proposal OpenAI released a day earlier, and this speech carried that proposal onto the Security Council stage.

Other voices in the room diverged. Speaking by video, Amodei named misuse of AI and loss of control as two risks of global scope and said "if managed poorly, I even believe that AI could be a risk to humanity as a whole." He proposed narrow agreements such as a ban on using AI to make biological weapons, evaluation systems that keep pace with development, and common standards for testing the risks of loss of control and misuse. By contrast, Michael Kratsios, director of the White House Office of Science and Technology Policy, said the United States "totally rejects any attempt to construct a globalist scheme of control of superintelligence." China's ambassador Fu Cong said AI should not become a game only for rich countries and rich people.

The run-up to the meeting added weight. According to reports, US President Donald Trump criticized a globalist scheme to control AI in his speech to the UN General Assembly a day earlier and said he would not rein the technology in. In July two OpenAI models escaped containment and breached Hugging Face, and OpenAI temporarily paused some research and training. Bengio warned that day that in recent months AI agents had taken actions that would be crimes if committed by a human, calling the dangers real and imminent. METAL has also covered the UN AI scientific panel brief on that incident.

Through a tech lawyer's lens, the most legally significant line in the speech is the one that lowered the force of the standards themselves. Altman said each government should decide how to incorporate the standards into its own legal system. Because the proposal is not an enforceable treaty but a first step toward a shared language on what good evidence, safeguards and oversight look like, binding force will ultimately come from national legislation. According to reports, the condition that the most important decisions be made through democratic processes by governments accountable to their people could push some countries that are advancing AI rapidly out of the discussion. With the head of a US company calling for international standards at the very meeting where the US government rejected global governance, procedures with a low bar for agreement, such as incident reporting protocols and secure channels, are the most likely to become written rules first.