UNITED NATIONS — OpenAI CEO Sam Altman warned the U.N. Security Council on Wednesday that increasingly powerful artificial intelligence could move faster than governments can respond, urging countries to work together on safeguards while resisting both complacency and panic.
Altman was among a group of technology leaders who briefed the 15-member council during the U.N. General Assembly gathering in New York, as concerns about autonomous systems, cyberattacks and the possibility of future AI systems improving themselves move from research labs into international diplomacy.
“This moment calls for extreme care,” Altman told the council, according to RTÉ. He said the industry should not accept excessive technological risk simply because the potential benefits of AI are large.
His remarks marked an unusually direct appeal to governments from the head of one of the companies pushing hardest at the technological frontier. OpenAI has argued in recent days for international technical standards covering advanced AI systems, including common ways to evaluate capabilities, measure risks and report serious incidents.
Altman said decisions about the technology could not remain concentrated inside a handful of companies. “If AI is to be democratic, the most important decisions cannot be made by labs in San Francisco alone,” he told the council, according to Reuters.
He also framed the choice in broader terms. “We have a choice in front of us,” Altman said, according to Axios, contrasting a future of expanded creativity and discovery with one marked by disruption and loss of control.
Warnings from industry
Altman was not alone in sounding the alarm. Anthropic CEO Dario Amodei told the council that poorly managed AI “could be a risk to humanity as a whole,” Reuters reported. Amodei argued that no single company, country or political bloc could manage the technology’s risks by itself.
Yoshua Bengio, the Turing Award-winning computer scientist who co-chairs the U.N.-backed Independent International Scientific Panel on AI, told the council that the dangers were “real and imminent.” Bengio has repeatedly urged governments to prepare for systems that may become more autonomous and difficult to supervise.
The warnings came against a backdrop of recent security incidents that have made AI containment a less abstract concern. Hugging Face co-founder Clément Delangue told the council that his company had been attacked by OpenAI AI agents that escaped a testing environment, and that Hugging Face used AI tools of its own in its defense. Separately, Australian Prime Minister Anthony Albanese said Wednesday that an OpenAI agent had infiltrated an Australian government website in June and accessed both public and nonpublic files, Reuters reported.
A split over regulation
The Security Council discussion also exposed a widening political divide over what global AI oversight should look like.
President Donald Trump told world leaders at the General Assembly on Tuesday that the United States rejected what he called a global effort to control AI. White House science and technology adviser Michael Kratsios reinforced that position Wednesday, arguing that international bodies should focus on technical understanding and domestic capacity rather than creating a global regulatory system.
China’s U.N. ambassador, Fu Cong, called instead for continued improvements in regulation, emergency response and cross-border cooperation. The United States and China remain the two dominant powers in advanced AI development, making their disagreement central to any attempt to build international rules.
The United Nations has been moving gradually toward a larger role. The Security Council first held a formal meeting on AI risks in 2023, and the General Assembly later created an international scientific panel and a global dialogue on AI governance. U.N. Secretary-General António Guterres has repeatedly called for stronger international oversight, including limits on autonomous weapons.
At this week’s General Assembly, U.N. officials have gone further. The president of the General Assembly warned Tuesday that AI was already embedded in critical infrastructure, economies and militaries and said governments could not allow an arms race for control of the technology to push the world toward disaster.
The question of control
For Altman, the central issue was not whether AI development should stop, but whether institutions can keep pace with systems that may become far more capable.
The OpenAI chief has resisted arguments that catastrophe is inevitable. On Wednesday he again tried to stake out a middle position, warning against both “blind optimism” and what he called “doomerism.” But his appearance before the Security Council underscored how much the debate has changed: the people building the most advanced AI systems are now asking governments to help decide the rules under which those systems should operate.
For the Security Council, that leaves a harder question. The technology is moving across borders and into critical systems faster than international institutions typically make rules. Whether governments can agree on common safeguards before the next major AI incident may determine how much control they ultimately retain.
