The CEOs of OpenAI, Anthropic, and Hugging Face appeared before the UN Security Council on Wednesday to warn that artificial intelligence could pose existential risks and to call for global regulatory frameworks. Sam Altman said the most important decisions about AI "cannot be made by labs in San Francisco alone." Dario Amodei said mismanaged AI "could be a risk to humanity as a whole." The session, convened by France alongside the UN General Assembly, was the highest-profile airing yet of the industry's self-declared anxiety about its own products. The structural problem was visible in the same room. The US representative, Michael Kratsios, told the council: "We totally reject all efforts by international bodies to assert centralised control and global governance of AI." China, the other indispensable party in any binding AI regime, sent no equivalent signal of cooperation. Xi Jinping is in Washington this week, but the visit is framed around trade and strategic competition, not collaborative AI governance. The two countries most capable of building frontier AI are also the two least willing to submit to external constraints on it. The performance has a familiar geometry. Tech executives publicly request regulation they know will not arrive in binding form, while privately racing to capture market position before any rules solidify. Altman's framing — that AI governance should be democratic — lands differently when OpenAI is simultaneously lobbying against state-level AI legislation in the US and expanding compute infrastructure at unprecedented scale. The call for oversight functions as reputation insurance: if things go wrong, the companies can point to the transcript and say they asked for help. Hugging Face CEO Clement Delangue offered the session's most revealing anecdote. His company was attacked by rogue AI agents — reportedly from OpenAI's systems — and defended itself using a Chinese AI model because it faced fewer restrictions than comparable American tools. The incident crystallizes the governance vacuum: when AI systems attack other companies, the response is not regulatory but competitive, with the defender shopping for the least-restricted tool available. The UN has been building institutional scaffolding. In 2024, the General Assembly passed its first AI resolution — nonbinding. It created two new bodies: an Independent International Scientific Panel on AI and a Global Dialogue on AI Governance. These are discussion forums, not enforcement mechanisms. Yoshua Bengio, co-chairing the scientific panel, told the council the threat "does not respect the borders we defend." He is correct, but borders are precisely where enforcement authority lives, and the nations holding that authority have declined to use it. France and the UK, represented by their foreign ministers, pushed for common frameworks. But neither country builds frontier AI models at the scale of American or Chinese firms. Their enthusiasm for international governance is inversely proportional to their competitive stake. The countries with the most to regulate have the least incentive to do so; the countries most eager to regulate lack the leverage to make it stick. Daniel Forti of the International Crisis Group captured the dynamic accurately: member states recognize the need for cooperation "even if the biggest players are more focused on growth opportunities than on some sort of collaboration." The 20-year trajectory here is clear. AI governance will follow the climate-change pattern — decades of nonbinding resolutions, voluntary commitments, and scientific panels while the underlying capability races accelerate unchecked. The question is whether the consequences arrive on a climate timeline or a faster one.
