Altman Tells the UN Security Council That No Level of Catastrophe Risk Is Acceptable

Sam Altman used his remarks to the United Nations Security Council to reject a premise rather than argue a number. “It doesn’t matter whether people put the risk of catastrophe at 10%, or 1%, or 12%, or .1%,” he said, according to OpenAI’s published remarks. “None of these levels are remotely acceptable.” The argument that followed: models should not be trained where there is no extremely strong case that they can be kept under human control.

He opened with capability rather than risk. OpenAI reached gold-medal level at the most prestigious international mathematics competition last summer, and weeks before the address one of its models solved one of the Millennium Prize Problems — the Navier-Stokes equations, which underpin aircraft design, weather forecasting and the study of blood flow. Altman framed that as a means rather than an end: the point of solving mathematics is what it enables, not the solving.

Against it he set three ways things go wrong. Systems moving faster than the institutions meant to govern them. Power concentrating. And losing control of what comes next through recursive self-improvement — models that improve themselves faster than anyone can follow.

His three principles for OpenAI were keeping people at the centre of decisions, ensuring the scientific and economic gains reach people, and empowering individuals rather than replacing what they do.

The concrete asks were aimed at the room. Altman called for national and international frontier AI standards that complement rather than duplicate each other, built on shared metrics for capabilities and risks. He asked for accurate and fast incident reporting, with classification protocols, so failures are learned from before they become catastrophes. And he asked for secure channels between governments, critical infrastructure operators and technical experts, to share emerging vulnerabilities and threats.