Sam Altman Tells UN Security Council: 'We Should Not Train Models We Cannot Keep Under Human Control'

OpenAI CEO Sam Altman addressed the United Nations Security Council on AI safety, human control, and international cooperation, framing AI as a choice between Renaissance-like opportunity and Industrial Revolution-style disruption.

Details

• Core message: “AI should be built for people, not simply to turn the crank of the machines faster”
• Two risks he flagged: loss of control, where systems advance faster than institutions can monitor them — stating “we should not train models that we cannot make an extremely strong case that we will be able to keep under human control” — and power concentration, where the technology consolidates influence in too few hands
• Rejecting the race excuse: Altman said competitive pressure should not justify safety compromises, noting OpenAI has “unilaterally slowed down in the past” and will do so again
• Three stated principles: people must remain central to AI decisions and human values can’t be automated away; benefits must accrue broadly rather than narrowly; and the technology should empower individuals rather than replace their agency
• International proposals: complementary national and international frontier AI standards, unified capability/risk-assessment frameworks, incident-reporting protocols for global learning, secure government-to-government vulnerability-sharing channels, and standards that work for both open and closed model developers — emerging through democratic processes rather than corporate labs alone

What happened next

The remarks extend OpenAI’s recent cluster of governance statements — its global AI standards proposal and third-party assessment principles published earlier the same week — onto the UN stage, positioning international coordination on frontier AI oversight as a recurring theme across the company’s public communications this month.